Every day, billions of people trust digital systems to run everything from communication to commerce to critical infrastructure. But the global early warning system that alerts security teams to dangerous software flaws is showing critical gaps in coverage—and most users have no idea their digital lives are likely becoming more vulnerable.
Over the past 18 months, two pillars of global cybersecurity have flirted with apparent collapse. In February 2024, the US-backed National Vulnerability Database (NVD)—relied on globally for its free analysis of security threats—abruptly stopped publishing new entries, citing a cryptic “change in interagency support.” Then, in April of this year, the Common Vulnerabilities and Exposures (CVE) program, the fundamental numbering system for tracking software flaws, seemed at similar risk: A leaked letter warned of an imminent contract expiration.
Cybersecurity practitioners have since flooded Discord channels and LinkedIn feeds with emergency posts and memes of “NVD” and “CVE” engraved on tombstones. Unpatched vulnerabilities are the second most common way cyberattackers break in, and they have led to fatal hospital outages and critical infrastructure failures. In a social media post, Jen Easterly, a US cybersecurity expert, said: “Losing [CVE] would be like tearing out the card catalog from every library at once—leaving defenders to sort through chaos while attackers take full advantage.” If CVEs identify each vulnerability like a book in a card catalogue, NVD entries provide the detailed review with context around severity, scope, and exploitability.
In the end, the Cybersecurity and Infrastructure Security Agency (CISA) extended funding for CVE another year, attributing the incident to a “contract administration issue.” But the NVD’s story has proved more complicated. Its parent organization, the National Institute of Standards and Technology (NIST), reportedly saw its budget cut roughly 12% in 2024, right around the time that CISA pulled its $3.7 million in annual funding for the NVD. Shortly after, as the backlog grew, CISA launched its own “Vulnrichment” program to help address the analysis gap, while promoting a more distributed approach that allows multiple authorized partners to publish enriched data.
“CISA continuously assesses how to most effectively allocate limited resources to help organizations reduce the risk of newly disclosed vulnerabilities,” says Sandy Radesky, the agency’s associate director for vulnerability management. Rather than just filling the gap, she emphasizes, Vulnrichment was established to provide unique additional information, like recommended actions for specific stakeholders, and to “reduce dependency of the federal government’s role to be the sole provider of vulnerability enrichment.”
Meanwhile, NIST has scrambled to hire contractors to help clear the backlog. Despite a return to pre-crisis processing levels, a boom in vulnerabilities newly disclosed to the NVD has outpaced these efforts. Currently, over 25,000 vulnerabilities await processing—nearly 10 times the previous high in 2017, according to data from the software company Anchore. Before that, the NVD largely kept pace with CVE publications, maintaining a minimal backlog.
“Things have been disruptive, and we’ve been going through times of change across the board,” Matthew Scholl, then chief of the computer security division in NIST’s Information Technology Laboratory, said at an industry event in April. “Leadership has assured me and everyone that NVD is and will continue to be a mission priority for NIST, both in resourcing and capabilities.” Scholl left NIST in May after 20 years at the agency, and NIST declined to comment on the backlog.
The situation has now prompted multiple government actions, with the Department of Commerce launching an audit of the NVD in May and House Democrats calling for a broader probe of both programs in June. But the damage to trust is already transforming geopolitics and supply chains as security teams prepare for a new era of cyber risk. “It’s left a bad taste, and people are realizing they can’t rely on this,” says Rose Gupta, who builds and runs enterprise vulnerability management programs. “Even if they get everything together tomorrow with a bigger budget, I don’t know that this won’t happen again. So I have to make sure I have other controls in place.”
As these public resources falter, organizations and governments are confronting a critical weakness in our digital infrastructure: Essential global cybersecurity services depend on a complex web of US agency interests and government funding that can be cut or redirected at any time.
Security haves and have-nots
What began as a trickle of software vulnerabilities in the early Internet era has become an unstoppable avalanche, and the free databases that have tracked them for decades have struggled to keep up. In early July, the CVE database crossed over 300,000 catalogued vulnerabilities. Numbers jump unpredictably each year, sometimes by 10% or much more. Even before its latest crisis, the NVD was notorious for delayed publication of new vulnerability analyses, often trailing private security software and vendor advisories by weeks or months.
Gupta has watched organizations increasingly adopt commercial vulnerability management (VM) software that includes its own threat intelligence services. “We’ve definitely become over-reliant on our VM tools,” she says, describing security teams’ growing dependence on vendors like Qualys, Rapid7, and Tenable to supplement or replace unreliable public databases. These platforms combine their own research with various data sources to create proprietary risk scores that help teams prioritize fixes. But not all organizations can afford to fill the NVD’s gap with premium security tools. “Smaller companies and startups, already at a disadvantage, are going to be more at risk,” she explains.
Komal Rawat, a security engineer in New Delhi whose mid-stage cloud startup has a limited budget, describes the impact in stark terms: “If NVD goes, there will be a crisis in the market. Other databases are not that popular, and to the extent they are adopted, they are not free. If you don’t have recent data, you’re exposed to attackers who do.”
The growing backlog means new devices could be more likely to have vulnerability blind spots—whether that’s a Ring doorbell at home or an office building’s “smart” access control system. The biggest risk may be “one-off” security flaws that fly under the radar. “There are thousands of vulnerabilities that will not affect the majority of enterprises,” says Gupta. “Those are the ones that we’re not getting analysis on, which would leave us at risk.”
NIST acknowledges it has limited visibility into which organizations are most affected by the backlog. “We don’t track which industries use which products and therefore cannot measure impact to specific industries,” a spokesperson says. Instead, the team prioritizes vulnerabilities on the basis of CISA’s known exploits list and those included in vendor advisories like Microsoft Patch Tuesday.
The biggest vulnerability
Brian Martin has watched this system evolve—and deteriorate—from the inside. A former CVE board member and an original project leader behind the Open Source Vulnerability Database, he has built a combative reputation over the decades as a leading historian and practitioner. Martin says his current project, VulnDB (part of Flashpoint Security), outperforms the official databases he once helped oversee. “Our team processes more vulnerabilities, at a much faster turnaround, and we do it for a fraction of the cost,” he says, referring to the tens of millions in government contracts that support the current system.
When we spoke in May, Martin said his database contains more than 112,000 vulnerabilities with no CVE identifiers—security flaws that exist in the wild but remain invisible to organizations that rely solely on public channels. “If you gave me the money to triple my team, that non-CVE number would be in the 500,000 range,” he said.
In the US, official vulnerability management duties are split between a web of contractors, agencies, and nonprofit centers like the Mitre Corporation. Critics like Martin saythat creates potential for redundancy, confusion, and inefficiency, with layers of middle management and relatively few actual vulnerability experts. Others defend the value of this fragmentation. “These programs build on or complement each other to create a more comprehensive, supportive, and diverse community,” CISA said in a statement. “That increases the resilience and usefulness of the entire ecosystem.”
As American leadership wavers, other nations are stepping up. China now operates multiple vulnerability databases, some surprisingly robust but tainted by the possibility that they are subject to state control. In May, the European Union accelerated the launch of its own database, as well as a decentralized “Global CVE” architecture. Following social media and cloud services, vulnerability intelligence has become another front in the contest for technological independence.
That leaves security professionals to navigate multiple potentially conflicting sources of data. “It’s going to be a mess, but I would rather have too much information than none at all,” says Gupta, describing how her team monitors multiple databases despite the added complexity.
Resetting software liability
As defenders adapt to the fragmenting landscape, the tech industry faces another reckoning: Why don’t software vendors carry more responsibility for protecting their customers from security issues? Major vendors routinely disclose—but don’t necessarily patch—thousands of new vulnerabilities each year. A single exposure could crash critical systems or increase the risks of fraud and data misuse.
For decades, the industry has hidden behind legal shields. “Shrink-wrap licenses” once forced consumers to broadly waive their right to hold software vendors liable for defects. Today’s end-user license agreements (EULAs), often delivered in pop-up browser windows, have evolved into incomprehensibly long documents. Last November, a lab project called “EULAS of Despair” used the length of War and Peace (587,287 words) to measure these sprawling contracts. The worst offender? Twitter, at 15.83 novels’ worth of fine print.
“This is a legal fiction that we’ve created around this whole ecosystem, and it’s just not sustainable,” says Andrea Matwyshyn, a US special advisor and technology law professor at Penn State University, where she directs the Policy Innovation Lab of Tomorrow. “Some people point to the fact that software can contain a mix of products and services, creating more complex facts. But just like in engineering or financial litigation, even the most messy scenarios can be resolved with the assistance of experts.”
This liability shield is finally beginning to crack. In July 2024, a faulty security update in CrowdStrike’s popular endpoint detection software crashed millions of Windows computers worldwide and caused outages at everything from airlines to hospitals to 911 systems. The incident led to billions in estimated damages, and the city of Portland, Oregon, even declared a “state of emergency.” Now, affected companies like Delta Airlines have hired high-priced attorneys to pursue major damages—a signal opening of the floodgates to litigation.
Despite the soaring number of vulnerabilities, many fall into long-established categories, such as SQL injections that interfere with database queries and buffer memory overflows that enable code to be executed remotely. Matwyshyn advocates for a mandatory “software bill of materials,” or S-BOM—an ingredients list that would let organizations understand what components and potential vulnerabilities exist throughout their software supply chains. One recent report found 30% of data breaches stemmed from the vulnerabilities of third-party software vendors or cloud service providers.
She adds: “When you can’t tell the difference between the companies that are cutting corners and a company that has really invested in doing right by their customers, that results in a market where everyone loses.”
CISA leadership shares this sentiment, with a spokesperson emphasizing its “secure-by-design principles,” such as “making essential security features available without additional cost, eliminating classes of vulnerabilities, and building products in a way that reduces the cybersecurity burden on customers.”
Avoiding a digital ‘dark age’
It will likely come as no surprise that practitioners are looking to AI to help fill the gap, while at the same time preparing for a coming swarm of cyberattacks by AI agents. Security researchers have used an OpenAI model to discover new “zero-day” vulnerabilities. And both the NVD and CVE teams are developing “AI-powered tools” to help streamline data collection, identification, and processing. NIST says that “up to 65% of our analysis time has been spent generating CPEs”—product information codes that pinpoint affected software. If AI can solve even part of this tedious process, it could dramatically speed up the analysis pipeline.
But Martin cautions against optimism around AI, noting that the technology remains unproven and often riddled with inaccuracies—which, in security, can be fatal. “Rather than AI or ML [machine learning], there are ways to strategically automate bits of the processing of that vulnerability data while ensuring 99.5% accuracy,” he says.
AI also fails to address more fundamental challenges in governance. The CVE Foundation, launched in April 2025 by breakaway board members, proposes a globally funded nonprofit model similar to that of the internet’s addressing system, which transitioned from US government control to international governance. Other security leaders are pushing to revitalize open-source alternatives like Google’s OSV Project or the NVD++ (maintained by VulnCheck), which are accessible to the public but currently have limited resources.
As these various reform efforts gain momentum, the world is waking up to the fact that vulnerability intelligence—like disease surveillance or aviation safety—requires sustained cooperation and public investment. Without it, a patchwork of paid databases will be all that remains, threatening to leave all but the richest organizations and nations permanently exposed.
Matthew King is a technology and environmental journalist based in New York. He previously worked for cybersecurity firm Tenable.
This week I’m sending congratulations to two sets of parents in South Africa. Babies Milayah and Rossouw arrived a few weeks ago. All babies are special, but these two set a new precedent. They’re the first to be born following “simplified” IVF performed in a mobile lab.
This new mobile lab is essentially a trailer crammed with everything an embryologist needs to perform IVF on a shoestring. It was designed to deliver reproductive treatments to people who live in rural parts of low-income countries, where IVF can be prohibitively expensive or even nonexistent. And it seems to work!
While IVF is increasingly commonplace in wealthy countries—around 12% of all births in Spain result from such procedures—it remains expensive and isn’t always covered by insurance or national health providers. And it’s even less accessible in low-income countries—especially for people who live in rural areas.
People often assume that countries with high birth rates don’t need access to fertility treatments, says Gerhard Boshoff, an embryologist at the University of Pretoria in South Africa. Sub-Saharan African countries like Niger, Angola, and Benin all have birth rates above 40 per 1,000 people, which is over four times the rates in Italy and Japan, for example.
For many people in rural areas of low-income countries, IVF clinics simply don’t exist. South Africa is considered a “reproductive hub” of the African continent, but even in that country there are fewer than 30 clinics for a population of over 60 million. A recent study found there were no such clinics in Angola or Malawi.
Willem Ombelet, a retired gynecologist, first noticed these disparities back in the 1980s, while he was working at an IVF lab in Pretoria. “I witnessed that infertility was [more prevalent] in the black population than the white population—but they couldn’t access IVF because of apartheid,” he says. The experience spurred him to find ways to make IVF accessible for everyone. In the 1990s, he launched The Walking Egg—a science and art project with that goal.
In 2008, Ombelet met Jonathan Van Blerkom, a reproductive biologist and embryologist who had already been experimenting with a simplified version of IVF. Typically, embryos are cultured in an incubator that provides a sterile mix of gases. Van Blerkom’s approach was to preload tubes with the required gases and seal them with a rubber stopper. “We don’t need a fancy lab,” says Ombelet.
Milayah was born on June 18.
COURTESY OF THE WALKING EGG
Eggs and sperm can be injected into the tubes through the stoppers, and the resulting embryos can be grown inside. All you really need is a good microscope and a way to keep the tube warm, says Ombelet. Once the embryos are around five days old, they can be transferred to a person’s uterus or frozen. “The cost is one tenth or one twentieth of a normal lab,” says Ombelet.
Ombelet, Van Blerkom, and their colleagues found that this approach appeared to work as well as regular IVF. The team ran their first pilot trial at a clinic in Belgium in 2012. The first babies conceived with the simplified IVF process were born later that year.
More recently, Boshoff wondered if the team could take the show on the road. Making IVF simpler and cheaper is one thing, but getting it to people who don’t have access to IVF care is another. What if the team could pack the simplified IVF lab into a trailer and drive it around rural South Africa?
“We just needed to figure out how to have everything in a very confined space,” says Boshoff. As part of the Walking Egg project, he and his colleagues found a way to organize the lab equipment and squeeze in air filters. He then designed a “fold-out system” that allowed the team to create a second room when the trailer was parked. This provides some privacy for people who are having embryos transferred, he says.
People who want to use the mobile IVF lab will first have to undergo treatment at a local medical facility, where they will take drugs that stimulate their ovaries to release eggs, and then have those eggs collected. The rest of the process can be done in the mobile lab, says Boshoff, who presented his work at the European Society of Human Reproduction and Embryology’s annual meeting in Paris earlier this month.
The first trial started last year. The team partnered with one of the few existing fertility clinics in rural South Africa, which put them in touch with 10 willing volunteers. Five of the 10 women got pregnant following their simplified IVF in the mobile lab. One miscarried, but four pregnancies continued. On June 18, baby Milayah arrived. Two days later, another mother welcomed baby Rossouw. The other babies could come any day now.
“We’ve proven that a very cheap and easy [IVF] method can be used even in a mobile unit and have comparable results to regular IVF,” says Ombelet, who says his team is planning similar trials in Egypt and Indonesia. “The next step is to roll it out all over the world.”
This article first appeared in The Checkup, MIT Technology Review’s weekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first, sign up here.
A new technique called LightShed will make it harder for artists to use existing protective tools to stop their work from being ingested for AI training. It’s the next step in a cat-and-mouse game—across technology, law, and culture—that has been going on between artists and AI proponents for years.
Generative AI models that create images need to be trained on a wide variety of visual material, and data sets that are used for this training allegedly include copyrighted art without permission. This has worried artists, who are concerned that the models will learn their style, mimic their work, and put them out of a job.
These artists got some potential defenses in 2023, when researchers created tools like Glaze and Nightshade to protect artwork by “poisoning” it against AI training (Shawn Shan was even named MIT Technology Review’s Innovator of the Year last year for his work on these). LightShed, however, claims to be able to subvert these tools and others like them, making it easy for the artwork to be used for training once again.
To be clear, the researchers behind LightShed aren’t trying to steal artists’ work. They just don’t want people to get a false sense of security. “You will not be sure if companies have methods to delete these poisons but will never tell you,” says Hanna Foerster, a PhD student at the University of Cambridge and the lead author of a paper on the work. And if they do, it may be too late to fix the problem.
AI models work, in part, by implicitly creating boundaries between what they perceive as different categories of images. Glaze and Nightshade change enough pixels to push a given piece of art over this boundary without affecting the image’s quality, causing the model to see it as something it’s not. These almost imperceptible changes are called perturbations, and they mess up the AI model’s ability to understand the artwork.
Glaze makes models misunderstand style (e.g., interpreting a photorealistic painting as a cartoon). Nightshade instead makes the model see the subject incorrectly (e.g., interpreting a cat in a drawing as a dog). Glaze is used to defend an artist’s individual style, whereas Nightshade is used to attack AI models that crawl the internet for art.
Foerster worked with a team of researchers from the Technical University of Darmstadt and the University of Texas at San Antonio to develop LightShed, which learns how to see where tools like Glaze and Nightshade splash this sort of digital poison onto art so that it can effectively clean it off. The group will present its findings at the Usenix Security Symposium, a leading global cybersecurity conference, in August.
The researchers trained LightShed by feeding it pieces of art with and without Nightshade, Glaze, and other similar programs applied. Foerster describes the process as teaching LightShed to reconstruct “just the poison on poisoned images.” Identifying a cutoff for how much poison will actually confuse an AI makes it easier to “wash” just the poison off.
LightShed is incredibly effective at this. While other researchers have found simple ways to subvert poisoning, LightShed appears to be more adaptable. It can even apply what it’s learned from one anti-AI tool—say, Nightshade—to others like Mist or MetaCloak without ever seeing them ahead of time. While it has some trouble performing against small doses of poison, those are less likely to kill the AI models’ abilities to understand the underlying art, making it a win-win for the AI—or a lose-lose for the artists using these tools.
Around 7.5 million people, many of them artists with small and medium-size followings and fewer resources, have downloaded Glaze to protect their art. Those using tools like Glaze see it as an important technical line of defense, especially when the state of regulation around AI training and copyright is still up in the air. The LightShed authors see their work as a warning that tools like Glaze are not permanent solutions. “It might need a few more rounds of trying to come up with better ideas for protection,” says Foerster.
The creators of Glaze and Nightshade seem to agree with that sentiment: The website for Nightshade warned the tool wasn’t future-proof before work on LightShed ever began. And Shan, who led research on both tools, still believes defenses like his have meaning even if there are ways around them.
“It’s a deterrent,” says Shan—a way to warn AI companies that artists are serious about their concerns. The goal, as he puts it, is to put up as many roadblocks as possible so that AI companies find it easier to just work with artists. He believes that “most artists kind of understand this is a temporary solution,” but that creating those obstacles against the unwanted use of their work is still valuable.
Foerster hopes to use what she learned through LightShed to build new defenses for artists, including clever watermarks that somehow persist with the artwork even after it’s gone through an AI model. While she doesn’t believe this will protect a work against AI forever, she thinks this could help tip the scales back in the artist’s favor once again.
China is the dominant force in next-generation energy technologies today. It’s pouring hundreds of billions of dollars into putting renewable sources like wind and solar on its grid, manufacturing millions of electric vehicles, and building out capacity for energy storage, nuclear power, and more. This investment has been transformational for the country’s economy and has contributed to establishing China as a major player in global politics.
Meanwhile, in the US, a massive new tax and spending bill just cut hundreds of billions in credits, grants, and loans for clean energy technologies. It’s a stark reversal from previous policies, and it could have massive effects at a time when it feels as if everyone is chasing China on energy.
So while we all try to get our heads around what’s next for climate tech in the US and beyond, let’s look at just how dominant China is when it comes to clean energy, as documented in three charts.
China is on an absolute tear installing wind and solar power. The country reached nearly 900 gigawatts of installed capacity for solar at the end of 2024, and the rapid pace of building has continued into this year. An additional 198 GW was installed between January and May, with 93 GW coming in May alone.
For context, those additions over the first five months of the year account for more than double the capacity of the grid in California. Not the renewables capacity of that state—the entire grid.
Meanwhile, the policy shift in the US is projected to slow down new solar and wind additions. With tax credits and other support stripped away, much of the new capacity that was expected to come online by the end of the decade will now face delays or cancellations.
That’s significant because of all the new electricity generation capacity that’s come online in the US recently, renewables make up the vast majority. Solar and battery storage alone are expected to make up over 80% of capacity additions in 2025. So slowing down wind and solar basically means slowing down adding new electricity capacity, at a time when demand is very much set to rise. (Hello, AI?)
China’s EV market is also booming—the country is currently flirting with a big symbolic milestone, nearing the point where over half of all new vehicles sold in the country are electric. (It already passed that mark for a single month and could do so on a yearly basis in the next couple of years.)
It’s not just selling those vehicles within China, either: the country exports them globally, with customers including established markets like Europe and growing ones like India and Brazil. As of 2024, more than 70% of electric and plug-in hybrid vehicles on roads around the world were built in ChinaSome leaders in legacy automakers are taking notice. Ford CEO Jim Farley shared some striking comments at the Aspen Ideas Festival last month about how far ahead China is on vehicle technology and price. “They have far superior in-vehicle technology,” Farley said. “We are in a global competition with China, and it’s not just EVs. And if we lose this, we do not have a future Ford.”
Looking ahead, China is still pouring money into renewables, storage, grids, and energy efficiency technologies. It’s also outspending the rest of the world on nuclear power. The country tripled its investment in renewable power from 2015 to 2025.
The situation isn’t set in stone, though: The US actually very briefly overtook China on battery investments over the past year, as Cat Clifford at Cipher reported last week. But changes resulting from the new bill could very quickly reverse that progress, cementing China as the place for battery manufacturing and innovation.
In a story earlier this week, the MIT economist David Autor laid out the high stakes for this race. Advanced manufacturing and technology are beneficial for US prosperity, and putting public support and trade protections in place for key industries could be crucial to keeping them going, he says.
I’d add that this whole discussion shouldn’t be about a zero-sum competition between the US and China. But manyexpertsargue that the US, where I and many readers live, is surrendering its leadership and ability to develop key energy technologies of the future.
Ultimately, the numbers don’t lie: By a lot of measures, China is the world’s leader in energy. The question is, will that change anytime soon?
This article is from The Spark, MIT Technology Review’s weekly climate newsletter. To receive it in your inbox every Wednesday,sign up here.
The “Big, Beautiful Bill” that President Donald Trump signed into law on July 4 was chock full of controversial policies—Medicaid work requirements, increased funding for ICE, and an end to tax credits for clean energy and vehicles, to name just a few. But one highly contested provision was missing. Just days earlier, during a late-night voting session, the Senate had killed the bill’s 10-year moratorium on state-level AI regulation.
“We really dodged a bullet,” says Scott Wiener, a California state senator and the author of SB 1047, a bill that would have made companies liable for harms caused by large AI models. It was vetoed by Governor Gavin Newsom last year, but Wiener is now working to pass SB 53, which establishes whistleblower protections for employees of AI companies. Had the federal AI regulation moratorium passed, he says, that bill likely would have been dead.
The moratorium could also have killed laws that have already been adopted around the country, including a Colorado law that targets algorithmic discrimination, laws in Utah and California aimed at making AI-generated content more identifiable, and other legislation focused on preserving data privacy and keeping children safe online. Proponents of the moratorium, such OpenAI and Senator Ted Cruz, have said that a “patchwork” of state-level regulations would place an undue burden on technology companies and stymie innovation. Federal regulation, they argue, is a better approach—but there is currently no federal AI regulation in place.
Wiener and other state lawmakers can now get back to work writing and passing AI policy, at least for the time being—with the tailwind of a major moral victory at their backs. The movement to defeat the moratorium was impressively bipartisan: 40 state attorneys general signed a letter to Congress opposing the measure, as did a group of over 250 Republican and Democratic state lawmakers. And while congressional Democrats were united against the moratorium, the final nail in its coffin was hammered in by Senator Marsha Blackburn of Tennessee, a Tea Party conservative and Trump ally who backed out of a compromise with Cruz at the eleventh hour.
The moratorium fight may have signaled a bigger political shift. “In the last few months, we’ve seen a much broader and more diverse coalition form in support of AI regulation generally,” says Amba Kak, co–executive director of the AI Now Institute. After years of relative inaction, politicians are getting concerned about the risks of unregulated artificial intelligence.
Granted, there’s an argument to be made that the moratorium’s defeat was highly contingent. Blackburn appears to have been motivated almost entirely by concerns about children’s online safety and the rights of country musicians to control their own likenesses; state lawmakers, meanwhile, were affronted by the federal government’s attempt to defang legislation that they had already passed.
And even though powerful technology firms such as Andreessen Horowitz and OpenAI reportedly lobbied in favor of the moratorium, continuing to push for it might not have been worth it to the Trump administration and its allies—at least not at the expense of tax breaks and entitlement cuts. Baobao Zhang, an associate professor of political science at Syracuse University, says that the administration may have been willing to give up on the moratorium in order to push through the rest of the bill by its self-imposed Independence Day deadline.
Andreessen Horowitz did not respond to a request for comment. OpenAI noted that the company was opposed to a state-by-state approach to AI regulation but did not respond to specific questions regarding the moratorium’s defeat.
It’s almost certainly the case that the moratorium’s breadth, as well as its decade-long duration, helped opponents marshall a diverse coalition to their side. But that breadth isn’t incidental—it’s related to the very nature of AI. Blackburn, who represents country musicians in Nashville, and Wiener, who represents software developers in San Francisco, have a shared interest in AI regulation precisely because such a powerful and general-purpose tool has the potential to affect so many people’s well-being and livelihood. “There are real anxieties that are touching people of all classes,” Kak says. “It’s creating solidarities that maybe didn’t exist before.”
Faced with outspoken advocates, concerned constituents, and the constant buzz of AI discourse, politicians from both sides of the aisle are starting to argue for taking AI extremely seriously. One of the most prominent anti-moratorium voices was Marjorie Taylor Greene, who voted for the version of the bill containing the moratorium before admitting that she hadn’t read it thoroughly and committing to opposing the moratorium moving forward. “We have no idea what AI will be capable of in the next 10 years,” she posted last month.
And two weeks ago, Pete Buttigieg, President Biden’s transportation secretary, published a Substack post entitled “We Are Still Underreacting on AI.” “The terms of what it is like to be a human are about to change in ways that rival the transformations of the Enlightenment or the Industrial Revolution, only much more quickly,” he wrote.
Wiener has noticed a shift among his peers. “More and more policymakers understand that we can’t just ignore this,” he says. But awareness is several steps short of effective legislation, and regulation opponents aren’t giving up the fight. The Trump administration is reportedly working on a slate of executive actions aimed at making more energy available for AI training and deployment, and Cruz says he is planning to introduce his own anti-regulation bill.
Meanwhile, proponents of regulation will need to figure out how to channel the broad opposition to the moratorium into support for specific policies. It won’t be a simple task. “It’s easy for all of us to agree on what we don’t want,” Kak says. “The harder question is: What is it that we do want?”
In a wide-ranging Roundtables conversation for MIT Technology Review subscribers, AI journalist and author Karen Hao spoke about her new book, Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI. She talked with executive editor Niall Firth about how she first covered the company in 2020 while on staff at MIT Technology Review, and they discussed how the AI industry now functions like an empire and what ethically-made AI looks like.
Read the transcript of the conversation, which has been lightly edited and condensed, below. Subscribers can watch the on-demand recording of the event here.
Niall Firth: Hello, everyone, and welcome to this special edition of Roundtables. These are our subscriber-only events where you get to listen in to conversations between editors and reporters. Now, I’m delighted to say we’ve got an absolute cracker of an event today. I’m very happy to have our prodigal daughter, Karen Hao, a fabulous AI journalist, here with us to talk about her new book. Hello, Karen, how are you doing?
Karen Hao: Good. Thank you so much for having me back, Niall.
Niall Firth: Lovely to have you. So I’m sure you all know Karen and that’s why you’re here. But to give you a quick, quick synopsis, Karen has a degree in mechanical engineering from MIT. She was MIT Technology Review’s senior editor for AI and has won countless awards, been cited in Congress, written for the Wall Street Journal and The Atlantic, and set up a series at the Pulitzer Center to teach journalists how to cover AI.
But most important of all, she’s here to discuss her new book, which I’ve got a copy of here, Empire of AI. The UK version is subtitled “Inside the reckless race for total domination,” and the US one, I believe, is “Dreams and nightmares in Sam Altman’s OpenAI.”
It’s been an absolute sensation, a New York Times chart topper. An incredible feat of reporting—like 300 interviews, including 90 with people inside OpenAI. And it’s a brilliant look at not just OpenAI’s rise, and the character of Sam Altman, which is very interesting in its own right, but also a really astute look at what kind of AI we’re building and who holds the keys.
Karen, the core of the book, the rise and rise of OpenAI, was one of your first big features at MIT Technology Review. It’s a brilliant story that lifted the lid for the first time on what was going on at OpenAI … and they really hated it, right?
Karen Hao: Yes, and first of all, thank you to everyone for being here. It’s always great to be home. I do still consider MIT Tech Review to be my journalistic home, and that story was—I only did it because Niall assigned it after I said, “Hey, it seems like OpenAI is kind of an interesting thing,” and he was like, you should profile them. And I had never written a profile about a company before, and I didn’t think that I would have it in me, and Niall believed that I would be able to do it. So it really didn’t happen other than because of you.
I went into the piece with an open mind about—let me understand what OpenAI is. Let me take what they say at face value. They were founded as a nonprofit. They have this mission to ensure artificial general intelligence benefits all of humanity. What do they mean by that? How are they trying to achieve that ultimately? How are they striking this balance between mission-driven AI development and the need to raise money and capital?
And through the course of embedding within the company for three days, and then interviewing dozens of people outside the company or around the company … I came to realize that there was a fundamental disconnect between what they were publicly espousing and accumulating a lot of goodwill from and how they were operating. And that is what I ended up focusing my profile on, and that is why they were not very pleased.
Niall Firth: And how have you seen OpenAI change even since you did the profile? That sort of misalignment feels like it’s got messier and more confusing in the years since.
Karen Hao: Absolutely. I mean, it’s kind of remarkable that OpenAI, you could argue that they are now one of the most capitalistic corporations in Silicon Valley. They just raised $40 billion, in the largest-ever private fundraising round in tech industry history. They’re valued at $300 billion. And yet they still say that they are first and foremost a nonprofit.
I think this really gets to the heart of how much OpenAI has tried to position and reposition itself throughout its decade-long history, to ultimately play into the narratives that they think are going to do best with the public and with policymakers, in spite of what they might actually be doing in terms of developing their technologies and commercializing them.
Niall Firth: You cite Sam Altman saying, you know, the race for AGI is what motivated a lot of this, and I’ll come back to that a bit before the end. But he talks about it as like the Manhattan Project for AI. You cite him quoting Oppenheimer (of course, you know, there’s no self-aggrandizing there): “Technology happens because it’s possible,” he says in the book.
And it feels to me like this is one of the themes of the book: the idea that technology doesn’t just happen because it comes along. It comes because of choices that people make. It’s not an inevitability that things are the way they are and that people are who they are. What they think is important—that influences the direction of travel. So what does this mean, in practice, if that’s the case?
Karen Hao: With OpenAI in particular, they made a very key decision early on in their history that led to all of the AI technologies that we see dominating the marketplace and dominating headlines today. And that was a decision to try and advance AI progress through scaling the existing techniques that were available to them. At the time when OpenAI started, at the end of 2015, and then, when they made that decision, in roughly around 2017, this was a very unpopular perspective within the broader AI research field.
There were kind of two competing ideas about how to advance AI progress, or rather a spectrum of ideas, bookended by two extremes. One extreme being, we have all the techniques we need, and we should just aggressively scale. And the other one being that we don’t actually have the techniques we need. We need to continue innovating and doing fundamental AI research to get more breakthroughs. And largely the field assumed that this side of the spectrum [focusing on fundamental AI research] was the most likely approach for getting advancements, but OpenAI was anomalously committed to the other extreme—this idea that we can just take neural networks and pump ever more data, and train on ever larger supercomputers, larger than have ever been built in history.
The reason why they made that decision was because they were competing against Google, which had a dominant monopoly on AI talent. And OpenAI knew that they didn’t necessarily have the ability to beat Google simply by trying to get research breakthroughs. That’s a very hard path. When you’re doing fundamental research, you never really know when the breakthrough might appear. It’s not a very linear line of progress, but scaling is sort of linear. As long as you just pump more data and more compute, you can get gains. And so they thought, we can just do this faster than anyone else. And that’s the way that we’re going to leap ahead of Google. And it particularly aligned with Sam Altman’s skillset, as well, because he is a once-in-a-generation fundraising talent, and when you’re going for scale to advance AI models, the primary bottleneck is capital.
And so it was kind of a great fit for what he had to offer, which is, he knows how to accumulate capital, and he knows how to accumulate it very quickly. So that is ultimately how you can see that technology is a product of human choices and human perspectives. And they’re the specific skills and strengths that that team had at the time for how they wanted to move forward.
Niall Firth: And to be fair, I mean, it works, right? It was amazing, fabulous. You know the breakthroughs that happened, GPT-2 to GPT-3, just from scale and data and compute, kind of were mind-blowing really, as we look back on it now.
Karen Hao: Yeah, it is remarkable how much it did work, because there was a lot of skepticism about the idea that scale could lead to the kind of technical progress that we’ve seen. But one of my biggest critiques of this particular approach is that there’s also an extraordinary amount of costs that come with this particular pathway to getting more advancements. And there are many different pathways to advancing AI, so we could have actually gotten all of these benefits, and moving forward, we could continue to get more benefits from AI, without actually engaging in a hugely consumptive, hugely costly approach to its development.
Niall Firth: Yeah, so in terms of consumptive, that’s something we’ve touched on here quite recently at MIT Technology Review, like the energy costs of AI. The data center costs are absolutely extraordinary, right? Like the data behind it is incredible. And it’s only gonna get worse in the next few years if we continue down this path, right?
Karen Hao: Yeah … so first of all, everyone should read the series that Tech Review put out, if you haven’t already, on the energy question, because it really does break down everything from what is the energy consumption of the smallest unit of interacting with these models, all the way up until the highest level.
The number that I have seen a lot, and that I’ve been repeating, is there was a McKinsey report that was looking at if we continue to just look at the pace at which data centers and supercomputers are being built and scaled, in the next five years, we would have to add two to six times the amount of energy consumed by California onto the grid. And most of that will have to be serviced by fossil fuels, because these data centers and supercomputers have to run 24/7, so we cannot rely solely on renewable energy. We do not have enough nuclear power capacity to power these colossal pieces of infrastructure. And so we’re already accelerating the climate crisis.
And we’re also accelerating a public-health crisis, the pumping of thousands of tons of air pollutants into the air from coal plants that are having their lives extended and methane gas turbines that are being built in service of powering these data centers. And in addition to that, there’s also an acceleration of the freshwater crisis, because these pieces of infrastructure have to be cooled with freshwater resources. It has to be fresh water, because if it’s any other type of water, it corrodes the equipment, it leads to bacterial growth.
And Bloomberg recently had a story that showed that two-thirds of these data centers are actually going into water-scarce areas, into places where the communities already do not have enough fresh water at their disposal. So that is one dimension of many that I refer to when I say, the extraordinary costs of this particular pathway for AI development.
Niall Firth: So in terms of costs and the extractive process of making AI, I wanted to give you the chance to talk about the other theme of the book, apart from just OpenAI’s explosion. It’s the colonial way of looking at the way AI is made: the empire. I’m saying this obviously because we’re here, but this is an idea that came out of reporting you started at MIT Technology Review and then continued into the book. Tell us about how this framing helps us understand how AI is made now.
Karen Hao: Yeah, so this was a framing that I started thinking a lot about when I was working on the AI Colonialism series for Tech Review. It was a series of stories that looked at the way that, pre-ChatGPT, the commercialization of AI and its deployment into the world was already leading to entrenchment of historical inequities into the present day.
And one example was a story that was about how facial recognition companies were swarming into South Africa to try and harvest more data from South Africa during a time when they were getting criticized for the fact that their technologies did not accurately recognize black faces. And the deployment of those facial recognition technologies into South Africa, into the streets of Johannesburg, was leading to what South African scholars were calling a recreation of a digital apartheid—the controlling of black bodies, movement of black people.
And this idea really haunted me for a really long time. Through my reporting in that series, there were so many examples that I kept hitting upon of this thesis, that the AI industry was perpetuating. It felt like it was becoming this neocolonial force. And then, when ChatGPT came out, it became clear that this was just accelerating.
When you accelerate the scale of these technologies, and you start training them on the entirety of the Internet, and you start using these supercomputers that are the size of dozens—if not hundreds—of football fields. Then you really start talking about an extraordinary global level of extraction and exploitation that is happening to produce these technologies. And then the historical power imbalances become even more obvious.
And so there are four parallels that I draw in my book between what I have now termed empires of AI versus empires of old. The first one is that empires lay claim to resources that are not their own. So these companies are scraping all this data that is not their own, taking all the intellectual property that is not their own.
The second is that empires exploit a lot of labor. So we see them moving to countries in the Global South or other economically vulnerable communities to contract workers to do some of the worst work in the development pipeline for producing these technologies—and also producing technologies that then inherently are labor-automating and engage in labor exploitation in and of themselves.
And the third feature is that the empires monopolize knowledge production. So, in the last 10 years, we’ve seen the AI industry monopolize more and more of the AI researchers in the world. So AI researchers are no longer contributing to open science, working in universities or independent institutions, and the effect on the research is what you would imagine would happen if most of the climate scientists in the world were being bankrolled by oil and gas companies. You would not be getting a clear picture, and we are not getting a clear picture, of the limitations of these technologies, or if there are better ways to develop these technologies.
And the fourth and final feature is that empires always engage in this aggressive race rhetoric, where there are good empires and evil empires. And they, the good empire, have to be strong enough to beat back the evil empire, and that is why they should have unfettered license to consume all of these resources and exploit all of this labor. And if the evil empire gets the technology first, humanity goes to hell. But if the good empire gets the technology first, they’ll civilize the world, and humanity gets to go to heaven. So on many different levels, like the empire theme, I felt like it was the most comprehensive way to name exactly how these companies operate, and exactly what their impacts are on the world.
Niall Firth: Yeah, brilliant. I mean, you talk about the evil empire. What happens if the evil empire gets it first? And what I mentioned at the top is AGI. For me, it’s almost like the extra character in the book all the way through. It’s sort of looming over everything, like the ghost at the feast, sort of saying like, this is the thing that motivates everything at OpenAI. This is the thing we’ve got to get to before anyone else gets to it.
There’s a bit in the book about how they’re talking internally at OpenAI, like, we’ve got to make sure that AGI is in US hands where it’s safe versus like anywhere else. And some of the international staff are openly like—that’s kind of a weird way to frame it, isn’t it? Why is the US version of AGI better than others?
So tell us a bit about how it drives what they do. And AGI isn’t an inevitable fact that’s just happening anyway, is it? It’s not even a thing yet.
Karen Hao: There’s not even consensus around whether or not it’s even possible or what it even is. There was recently a New York Times story by Cade Metz that was citing a survey of long-standing AI researchers in the field, and 75% of them still think that we don’t have the techniques yet for reaching AGI, whatever that means. And the most classic definition or understanding of what AGI is, is being able to fully recreate human intelligence in software. But the problem is, we also don’t have scientific consensus around what human intelligence is. And so one of the aspects that I talk about a lot in the book is that, when there is a vacuum of shared meaning around this term, and what it would look like, when would we have arrived at it? What capabilities should we be evaluating these systems on to determine that we’ve gotten there? It can basically just be whatever OpenAI wants.
So it’s kind of just this ever-present goalpost that keeps shifting, depending on where the company wants to go. You know, they have a full range, a variety of different definitions that they’ve used throughout the years. In fact, they even have a joke internally: If you ask 13 OpenAI researchers what AGI is, you’ll get 15 definitions. So they are kind of self-aware that this is not really a real term and it doesn’t really have that much meaning.
But it does serve this purpose of creating a kind of quasi-religious fervor around what they’re doing, where people think that they have to keep driving towards this horizon, and that one day when they get there, it’s going to have a civilizationally transformative impact. And therefore, what else should you be working on in your life, but this? And who else should be working on it, but you?
And so it is their justification not just for continuing to push and scale and consume all these resources—because none of that consumption, none of that harm matters anymore if you end up hitting this destination. But they also use it as a way to develop their technologies in a very deeply anti-democratic way, where they say, we are the only people that have the expertise, that have the right to carefully control the development of this technology and usher it into the world. And we cannot let anyone else participate because it’s just too powerful of a technology.
Niall Firth: You talk about the factions, particularly the religious framing. AGI has been around as a concept for a while—it was very niche, very kind of nerdy fun, really, to talk about—to suddenly become extremely mainstream. And they have the boomers versus doomers dichotomy. Where are you on that spectrum?
Karen Hao: So the boomers are people who think that AGI is going to bring us to utopia, and the doomers think AGI is going to devastate all of humanity. And to me these are actually two sides of the same coin. They both believe that AGI is possible, and it’s imminent, and it’s going to change everything.
And I am not on this spectrum. I’m in a third space, which is the AI accountability space, which is rooted in the observation that these companies have accumulated an extraordinary amount of power, both economic and political power, to go back to the empire analogy.
Ultimately, the thing that we need to do in order to not return to an age of empire and erode a lot of democratic norms is to hold these companies accountable with all the tools at our disposal, and to recognize all the harms that they are already perpetuating through a misguided approach to AI development.
Niall Firth: I’ve got a couple of questions from readers. I’m gonna try to pull them together a little bit because Abbas asks, what would post-imperial AI look like? And there was a question from Liam basically along the same lines. How do you make a more ethical version of AI that is not within this framework?
Karen Hao: We sort of already touched a little bit upon this idea. But there are so many different ways to develop AI. There are myriads of techniques throughout the history of AI development, which is decades long. There have been various shifts in the winds of which techniques ultimately rise and fall. And it isn’t based solely on the scientific or technical merit of any particular technique. Oftentimes certain techniques become more popular because of business reasons or because of the funder’s ideologies. And that’s sort of what we’re seeing today with the complete indexing of AI development on large-scale AI model development.
And ultimately, these large-scale models … We talked about how it’s a remarkable technical leap, but in terms of social progress or economic progress, the benefits of these models have been kind of middling. And the way that I see us shifting to AI models that are going to be A) more beneficial and B) not so imperial is to refocus on task-specific AI systems that are tackling well-scoped challenges that inherently lend themselves to the strengths of AI systems that are inherently computational optimization problems.
So I’m talking about things like using AI to integrate more renewable energy into the grid. This is something that we definitely need. We need to more quickly accelerate our electrification of the grid, and one of the challenges of using more renewable energy is the unpredictability of it. And this is a key strength of AI technologies, being able to have predictive capabilities and optimization capabilities where you can match the energy generation of different renewables with the energy demands of different people that are drawing from the grid.
Niall Firth: Quite a few people have been asking, in the chat, different versions of the same question. If you were an early-career AI scientist, or if you were involved in AI, what can you do yourself to bring about a more ethical version of AI? Do you have any power left, or is it too late?
Karen Hao: No, I don’t think it’s too late at all. I mean, as I’ve been talking with a lot of people just in the lay public, one of the biggest challenges that they have is they don’t have any alternatives for AI. They want the benefits of AI, but they also do not want to participate in a supply chain that is really harmful. And so the first question is, always, is there an alternative? Which tools do I shift to? And unfortunately, there just aren’t that many alternatives right now.
And so the first thing that I would say to early-career AI researchers and entrepreneurs is to build those alternatives, because there are plenty of people that are actually really excited about the possibility of switching to more ethical alternatives. And one of the analogies I often use is that we kind of need to do with the AI industry what happened with the fashion industry. There was also a lot of environmental exploitation, labor exploitation in the fashion industry, and there was enough consumer demand that it created new markets for ethical and sustainably sourced fashion. And so we kind of need to see just more options occupying that space.
Niall Firth: Do you feel optimistic about the future? Or where do you sit? You know, things aren’t great as you spell them out now. Where’s the hope for us?
Karen Hao: I am. I’m super optimistic. Part of the reason why I’m optimistic is because you know, a few years ago, when I started writing about AI at Tech Review, I remember people would say, wow, that’s a really niche beat. Do you have enough to write about?
And now, I mean, everyone is talking about AI, and I think that’s the first step to actually getting to a better place with AI development. The amount of public awareness and attention and scrutiny that is now going into how we develop these technologies, how we use these technologies, is really, really important. Like, we need to be having this public debate and that in and of itself is a significant step change from what we had before.
But the next step, and part of the reason why I wrote this book, is we need to convert the awareness into action, and people should take an active role. Every single person should feel that they have an active role in shaping the future of AI development, if you think about all of the different ways that you interface with the AI development supply chain and deployment supply chain—like you give your data or withhold your data.
There are probably data centers that are being built around you right now. If you’re a parent, there’s some kind of AI policy being crafted at [your kid’s] school. There’s some kind of AI policy being crafted at your workplace. These are all what I consider sites of democratic contestation, where you can use those opportunities to assert your voice about how you want AI to be developed and deployed. If you do not want these companies to use certain kinds of data, push back when they just take the data.
I closed all of my personal social media accounts because I just did not like the fact that they were scraping my personal photos to train their generative AI models. I’ve seen parents and students and teachers start forming committees within schools to talk about what their AI policy should be and to draft it collectively as a community. Same with businesses. They’re doing the same thing. If we all kind of step up to play that active role, I am super optimistic that we’ll get to a better place.
Niall Firth: Mark, in the chat, mentions the Māori story from New Zealand towards the end of your book, and that’s an example of sort of community-led AI in action, isn’t it?
Karen Hao: Yeah. There was a community in New Zealand that really wanted to help revitalize the Māori language by building a speech recognition tool that could recognize Māori, and therefore be able to transcribe a rich repository of archival audio of their ancestors speaking Māori. And the first thing that they did when engaging in that project was they asked the community, do you want this AI tool?
Niall Firth: Imagine that.
Karen Hao: I know! It’s such a radical concept, this idea of consent at every stage. But they first asked that; the community wholeheartedly said yes. They then engaged in a public education campaign to explain to people, okay, what does it take to develop an AI tool? Well, we are going to need data. We’re going to need audio transcription pairs to train this AI model. So then they ran a public contest in which they were able to get dozens, if not hundreds, of people in their community to donate data to this project. And then they made sure that when they developed the model, they actively explained to the community at every step how their data was being used, how it would be stored, how it would continue to be protected. And any other project that would use the data has to get permission and consent from the community first.
And so it was a completely democratic process, for whether they wanted the tool, how to develop the tool, and how the tool should continue to be used, and how their data should continue to be used over time.
Niall Firth: Great. I know we’ve gone a bit over time. I’ve got two more things I’m going to ask you, basically putting together lots of questions people have asked in the chat about your view on what role regulations should play. What are your thoughts on that?
Karen Hao: Yeah, I mean, in an ideal world where we actually had a functioning government, regulation should absolutely play a huge role. And it shouldn’t just be thinking about once an AI model is built, how to regulate that. But still thinking about the full supply chain of AI development, regulating the data and what’s allowed to be trained in these models, regulating the land use. And what pieces of land are allowed to build data centers? How much energy and water are the data centers allowed to consume? And also regulating the transparency. We don’t know what data is in these training data sets, and we don’t know the environmental costs of training these models. We don’t know how much water these data centers consume and that is all information that these companies actively withhold to prevent democratic processes from happening. So if there were one major intervention that regulators could have, it should be to dramatically increase the amount of transparency along the supply chain.
Niall Firth: Okay, great. So just to bring it back around to OpenAI and Sam Altman to finish with. He famously sent an email around, didn’t he? After your original Tech Review story, saying this is not great. We don’t like this. And he didn’t want to speak to you for your book, either, did he?
Karen Hao: No, he did not.
Niall Firth: No. But imagine Sam Altman is in the chat here. He’s subscribed to Technology Review and is watching this Roundtables because he wants to know what you’re saying about him. If you could talk to him directly, what would you like to ask him?
Karen Hao: What degree of harm do you need to see in order to realize that you should take a different path?
Niall Firth: Nice, blunt, to the point. All right, Karen, thank you so much for your time.
Karen Hao: Thank you so much, everyone.
MIT Technology Review Roundtables is a subscriber-only online event series where experts discuss the latest developments and what’s next in emerging technologies. Sign up to get notified about upcoming sessions.
If you were told that the odds of something were 3.1%, it really wouldn’t seem like much. But for the people charged with protecting our planet, it was huge.
On February 18, astronomers determined that a 130- to 300-foot-long asteroid had a 3.1% chance of crashing into Earth in 2032. Never had an asteroid of such dangerous dimensions stood such a high chance of striking the planet. For those following this developing story in the news, the revelation was unnerving. For many scientists and engineers, though, it turned out to be—despite its seriousness—a little bit exciting.
While possible impact locations included patches of empty ocean, the space rock, called 2024 YR4, also had several densely populated cities in its possible crosshairs, including Mumbai, Lagos, and Bogotá. If the asteroid did in fact hit such a metropolis, the best-case scenario was severe damage; the worst case was outright, total ruin. And for the first time, a group of United Nations–backed researchers began to have high-level discussions about the fate of the world: If this asteroid was going to hit the planet, what sort of spaceflight mission might be able to stop it? Would they ram a spacecraft into it to deflect it? Would they use nuclear weapons to try to swat it away or obliterate it completely?
At the same time, planetary defenders all over the world crewed their battle stations to see if we could avoid that fate—and despite the sometimes taxing new demands on their psyches and schedules, they remained some of the coolest customers in the galaxy. “I’ve had to cancel an appointment saying, I cannot come—I have to save the planet,” says Olivier Hainaut, an astronomer at the European Southern Observatory and one of those who tracked down 2024 YR4.
Then, just as quickly as history was made, experts declared that the danger had passed. On February 24, asteroid trackers issued the all-clear: Earth would be spared, just as many planetary defense researchers had felt assured it would.
How did they do it? What was it like to track the rising (and rising and rising) danger of this asteroid, and to ultimately determine that it’d miss us?
This is the inside story of how, over a span of just two months, a sprawling network of global astronomers found, followed, mapped, planned for, and finally dismissed 2024 YR4, the most dangerous asteroid ever found—all under the tightest of timelines and, for just a moment, with the highest of stakes.
“It was not an exercise,” says Hainaut. This was the real thing: “We really [had] to get it right.”
IN THE BEGINNING
December 27, 2024
THE ASTEROID TERRESTRIAL-IMPACT LAST ALERT SYSTEM, HAWAII
Long ago, an asteroid in the space-rock highway between Mars and Jupiter felt a disturbance in the force: the gravitational pull of Jupiter itself, king of the planets. After some wobbling back and forth, this asteroid was thrown out of the belt, skipped around the sun, and found itself on an orbit that overlapped with Earth’s own.
“I was the first one to see the detections of it,” Larry Denneau, of the University of Hawai‘i, recalls. “A tiny white pixel on a black background.”
Denneau is one of the principal investigators at the NASA-funded Asteroid Terrestrial-impact Last Alert System (ATLAS) telescopic network. It may have been just two days after Christmas, but he followed procedure as if it were any other day of the year and sent the observations of the tiny pixel onward to another NASA-funded facility, the Minor Planet Center (MPC) in Cambridge, Massachusetts.
There’s an alternate reality in which none of this happened. Fortunately, in our timeline, various space agencies—chiefly NASA, but also the European Space Agency and the Japan Aerospace Exploration Agency—invest millions of dollars every year in asteroid-spotting efforts.
And while multiple nations host observatories capable of performing this work, the US clearly leads the way: Its planetary defense program provides funding to a suite of telescopic facilities solely dedicated to identifying potentially hazardous space rocks. (At least, it leads the way for the moment. The White House’s proposal for draconian budget cuts to NASA and the National Science Foundation mean that several observatories and space missions linked to planetary defense are facing funding losses or outright terminations.)
Astronomers working at these observatories are tasked with finding threatening asteroids before they find us—because you can’t fight what you can’t see. “They are the first line of planetary defense,” says Kelly Fast, the acting planetary defense officer at NASA’s Planetary Defense Coordination Office in Washington, DC.
ATLAS is one part of this skywatching project, and it consists of four telescopes: two in Hawaii, one in Chile, and another in South Africa. They don’t operate the way you’d think, with astronomers peering through them all night. Instead, they operate “completely robotically and automatically,” says Denneau. Driven by coding scripts that he and his colleagues have developed, these mechanical eyes work in harmony to watch out for any suspicious space rocks. Astronomers usually monitor their survey of the sky from a remote location.
ATLAS telescopes are small, so they can’t see particularly distant objects. But they have a wide field of view, allowing them to see large patches of space at any one moment. “As long as the weather is good, we’re constantly monitoring the night sky, from the North Pole to the South Pole,” says Denneau.
Larry Denneau is a principal investigator at the Asteroid Terrestrial-impact Last Alert System telescopic network.
COURTESY PHOTO
If they detect the starlight reflecting off a moving object, an operator, such as Denneau, gets an alert and visually verifies that the object is real and not some sort of imaging artifact. When a suspected asteroid (or comet) is identified, the observations are sent to the MPC, which is home to a bulletin board featuring (among other things) orbital data on all known asteroids and comets.
If the object isn’t already listed, a new discovery is announced, and other astronomers can perform follow-up observations.
In just the past few years, ATLAS has detected more than 1,200 asteroids with near-Earth orbits. Finding ultimately harmless space rocks is routine work—so much so that when the new near-Earth asteroid was spotted by ATLAS’s Chilean telescope that December day, it didn’t even raise any eyebrows.
Denneau had simply been sitting at home,doing some late-night work on his computer. At the time, of course, he didn’t know that his telescope had just spied what would soon become a history-making asteroid—one that could alter the future of the planet.
The MPC quickly confirmed the new space rock hadn’t already been “found,” and astronomers gave it a provisional designation: 2024 YR4.
CATALINA SKY SURVEY, ARIZONA
Around the same time, the discovery was shared with another NASA-funded facility: the Catalina Sky Survey, a nest of three telescopes in the Santa Catalina Mountains north of Tucson that works out of the University of Arizona. “We run a very tight operation,” says Kacper Wierzchoś, one of its comet and asteroid spotters. Unlike ATLAS, these telescopes (although aided by automation) often have an in-person astronomer available to quickly alter the surveys in real time.
“We run a very tight operation,” says Kacper Wierzchoś, one of the comet and asteroid spotters at the Catalina Sky Survey north of Tucson, Arizona.
COURTESY PHOTO
So when Catalina was alerted about what its peers at ATLAS had spotted, staff deployed its Schmidt telescope—a smaller one that excels at seeing bright objects moving extremely quickly. As they fed their own observations of 2024 YR4 to the MPC, Catalina engineer David Rankin looked back over imagery from the previous days and found the new asteroid lurking in a night-sky image taken on December 26. Around then, ATLAS also realized that it had caught sight of 2024 YR4 in a photograph from December 25.
The combined observations confirmed it: The asteroid had made its closest approach to Earth on Christmas Day, meaning it was already heading back out into space. But where, exactly, was this space rock going? Where would it end up after it swung around the sun?
CENTER FOR NEAR-EARTH OBJECT STUDIES, CALIFORNIA
If the answer to that question was Earth, Davide Farnocchia would be one of the first to know. You could say he’s one of NASA’s watchers on the wall.
And he’s remarkably calm about his duties. When he first heard about 2024 YR4, he barely flinched. It was just another asteroid drifting through space not terribly far from Earth. It was another box to be ticked.
Once it was logged by the MPC, it was Farnocchia’s job to try to plot out 2024 YR4’s possible paths through space, checking to see if any of them overlapped with our planet’s. He works at NASA’s Center for Near-Earth Object Studies (CNEOS) in California, where he’s partly responsible for keeping track of all the known asteroids and comets in the solar system. “We have 1.4 million objects to deal with,” he says, matter-of-factly.
In the past, astronomers would have had to stitch together multiple images of this asteroid and plot out its possible trajectories. Today, fortunately, Farnocchia has some help: He oversees the digital brain Sentry, an autonomous system he helped code. (Two other facilities in Italy perform similar work: the European Space Agency’s Near-Earth Object Coordination Centre, or NEOCC, and the privately owned Near-Earth Objects Dynamics Site, or NEODyS.)
To chart their courses, Sentry uses every new observation of every known asteroid or comet listed on the MPC to continuously refine the orbits of all those objects, using the immutable laws of gravity and the gravitational influences of any planets, moons, or other sizable asteroids they pass. A recent update to the software means that even the ever-so-gentle push afforded by sunlight is accounted for. That allows Sentry to confidently project the motions of all these objects at least a century into the future.
Davide Farnocchia helps track all the known asteroids and comets in the solar system at NASA’s Center for Near-Earth Object Studies.
COURTESY PHOTO
Almost all newly discovered asteroids are quickly found to pose no impact risk. But those that stand even an infinitesimally small chance of smashing into our planet within the next 100 years are placed on the Sentry Risk List until additional observations can rule out those awful possibilities. Better safe than sorry.
In late December, with just a limited set of data, Sentry concluded that there was a non-negligible chance 2024 YR4 would strike Earth in 2032. Aegis, the equivalent software at Europe’s NEOCC site, agreed. No bother. More observations would very likely remove 2024 YR4 from the Risk List. Just another day at the office for Farnocchia.
It’s worth noting that an asteroid heading toward Earth isn’t always a problem. Small rocks burn up in the planet’s atmosphere several times a day; you’ve probably seen one already this year, on a moonless night. But above a certain size, these rocks turn from innocuous shooting stars into nuclear-esque explosions.
Reflected starlight is great for initially spotting asteroids, but it’s a terrible way to determine how big they are. A large, dull rock reflects as much light as a bright, tiny rock, making them appear the same to many telescopes. And that’s a problem, considering that a rock around 30 feet long will explode loudly but inconsequentially in Earth’s atmosphere, while a 3,000-foot-long asteroid would slam into the ground and cause devastation on a global scale, imperiling all of civilization. Roughly speaking, if you double the size of an asteroid, it becomes eight times more energetic upon impact—so finding out the size of an Earthbound asteroid is of paramount importance.
In those first few hours after it was discovered, and before anyone knew how shiny or dull its surface was, 2024 YR4 was estimated by astronomers to be as small as 65 feet across or as large as 500 feet. An object of the former size would blow up in mid-air, shattering windows over many miles and likely injuring thousands of people. At the latter size it would vaporize the heart of any city it struck, turning solid rock and metal into liquid and vapor, while its blast wave would devastate the rest of it, killing hundreds of thousands or even millions in the process.
So now the question was: Just how big was 2024 YR4?
REFINING THE PICTURE
Mid-January 2025
VERY LARGE TELESCOPE, CHILE
Understandably dissatisfied with that level of imprecision, the European Southern Observatory’s Very Large Telescope (VLT), high up on the Cerro Paranal mountain in Chile’s Atacama Desert, entered the chat. As the name suggests, this flagship facility is vast, and it’s capable of really zooming in on distant objects. Or to put it another way: “The VLT is the largest, biggest, best telescope in the world,” says Hainaut, one of the facility’s operators, who usually commands it from half a world away in Germany.
In reality, the VLT—which lends a hand to the European Space Agency in its asteroid-hunting duties—is actually made up of four massive telescopes, each fixed on four separate corners of the sky. They can be combined to act as a huge light bucket, allowing astronomers to see very faint asteroids. Four additional, smaller, movable telescopes can also team up with their bigger siblings to provide remarkably high-resolution images of even the stealthiest space rocks.
In this sequence of infrared images taken by ESO’s VLT, the individual image frames have been aligned so that the asteroid remains in the center as other stars appear to move around it.
ESO/O. HAINAUT ET AL.
With so much tech to oversee, the control room of the VLT looks a bit like the inside of the Death Star. “You have eight consoles, each of them with a dozen screens. It’s big, it’s large, it’s spectacular,” says Hainaut.
In mid-January, the European Space Agency asked the VLT to study several asteroids that had somewhat suspicious near-Earth orbits—including 2024 YR4. With just a few lines of code, the VLT could easily train its sharp eyes on an asteroid like 2024 YR4, allowing astronomers to narrow down its size range. It was found to be at least 130 feet long (big enough to cause major damage in a city) and as much as 300 feet (able to annihilate one).
January 29, 2025
INTERNATIONAL ASTEROID WARNING NETWORK
Marco Fenucci is a near-Earth-object dynamicist at the European Space Agency’s Near-Earth Object Coordination Centre.
COURTESY PHOTO
By the end of the month, there was no mistaking it: 2024 YR4 stood a greater than 1% chance of impacting Earth on December 22, 2032.
“It’s not something you see very often,” says Marco Fenucci, a near-Earth-object dynamicist at NEOCC. He admits that although it was “a serious thing,” this escalation was also “exciting to see”—something straight out of a sci-fi flick.
Sentry and Aegis, along with the systems at NEODyS, had been checking one another’s calculations. “There was a lot of care,” says Farnocchia, who explains that even though their programs worked wonders, their predictions were manually verified by multiple experts. When a rarity like 2024 YR4 comes along, he says, “you kind of switch gears, and you start being more cautious. You start screening everything that comes in.”
At this point, the klaxon emanating from these three data centers pushed the International Asteroid Warning Network (IAWN), a UN-backed planetary defense awareness group, to issue a public alert to the world’s governments: The planet may be in peril. For the most part, it was at this moment that the media—and the wider public—became aware of the threat. Earth, we may have a problem.
Denneau, along with plenty of other astronomers, received an urgent email from Fast at NASA’s Planetary Defense Coordination Office, requesting that all capable observatories track this hazardous asteroid. But there was one glaring problem. When 2024 YR4 was discovered on December 27, it was already two days after it had made its closest approach to Earth. And since it was heading back out into the shadows of space, it was quickly fading from sight.
Once it gets too faint, “there’s not much ATLAS can do,” Denneau says. By the time of IAWN’s warning, planetary defenders had just weeks to try to track 2024 YR4 and refine the odds of its hitting Earth before they’d lose it to the darkness.
And if their scopes failed, the odds of an Earth impact would have stayed uncomfortably high until 2028, when the asteroid was due to make another flyby of the planet. That’d be just four short years before the space rock might actually hit.
“In that situation, we would have been … in trouble,” says NEOCC’s Fenucci.
The hunt was on.
PREPARING FOR THE WORST
February 5 and February 6, 2025
SPACE MISSION PLANNING ADVISORY GROUP, AUSTRIA
In early February, spaceflight mission specialists, including those at the UN-supported Space Mission Planning Advisory Group in Vienna, began high-level talks designed to sketch out ways in which 2024 YR4 could be either deflected away from Earth or obliterated—you know, just in case.
A range of options were available—including ramming it with several uncrewed spacecraft or assaulting it with nuclear weapons—but there was no silver bullet in this situation. Nobody had ever launched a nuclear explosive device into deep space before, and the geopolitical ramifications of any nuclear-armed nations doing so in the present day would prove deeply unwelcome. Asteroids are also extremely odd objects; some, perhaps including 2024 YR4, are less like single chunks of rock and more akin to multiple cliffs flying in formation. Hit an asteroid like that too hard and you could fail to deflect it—and instead turn an Earthbound cannonball into a spray of shotgun pellets.
It’s safe to say that early on, experts were concerned about whether they could prevent a potential disaster. Crucially, eight years was not actually much time to plan something of this scale. So they were keen to better pinpoint how likely, or unlikely, it was that 2024 YR4 was going to collide with the planet before any complex space mission planning began in earnest.
The people involved with these talks—from physicists at some of America’s most secretive nuclear weapons research laboratories to spaceflight researchers over in Europe—were not feeling close to anything resembling panic. But “the timeline was really short,” admits Hainaut. So there was an unprecedented tempo to their discussions. This wasn’t a drill. This was the real deal. What would they do to defend the planet if an asteroid impact couldn’t be ruled out?
Luckily, over the next few days, a handful of new observations came in. Each helped Sentry, Aegis, and the system at NEODyS rule out more of 2024 YR4’s possible future orbits. Unluckily, Earth remained a potential port of call for this pesky asteroid—and over time, our planet made up a higher proportion of those remaining possibilities. That meant that the odds of an Earth impact “started bubbling up,” says Denneau.
EVA REDAMONTI
By February 6, they jumped to 2.3%—a one-in-43 chance of an impact.
“How much anxiety someone should feel over that—it’s hard to say,” Denneau says, with a slight shrug.
In the past, several elephantine asteroids have been found to stand a small chance of careening unceremoniously into the planet. Such incidents tend to follow a pattern. As more observations come in and the asteroid’s orbit becomes better known, an Earth impact trajectory remains a possibility while other outlying orbits are removed from the calculations—so for a time, the odds of an impact rise. Finally, with enough observations in hand, it becomes clear that the space rock will miss our world entirely, and the impact odds plummet to zero.
Astronomers expected this to repeat itself with 2024 YR4. But there was no guarantee. There’s no escaping the fact that one day, sooner or later, scientists will discover a dangerous asteroid that will punch Earth in the face—and raze a city in the process.
After all, asteroids capable of trashing a city have found their way to Earth plenty of times before, and not just in the very distant past. In 1908, an 800-square-mile patch of forest in Siberia—one that was, fortunately, very sparsely populated—was decimated by a space rock just 180 feet long. It didn’t even hit the ground; it exploded in midair with the force of a 15-megaton blast.
But only one other asteroid comparable in size to 2024 YR4 had its 2.3% figure beat: in 2004, Apophis—capable of causing continental-scale damage—had (briefly) stood a 2.7% chance of impacting Earth in 2029.
Rapidly approaching uncharted waters, the powers that be at NASA decided to play a space-based wild card: the James Webb Space Telescope, or JWST.
THE JAMES WEBB SPACE TELESCOPE, DEEP SPACE, ONE MILLION MILES FROM EARTH
A large dull asteroid reflects the same amount of light as a small shiny one, but that doesn’t mean astronomers sizing up an asteroid are helpless. If you view both asteroids in the infrared, the larger one glows brighter than the smaller one no matter the surface coating—making infrared, or the thermal part of the electromagnetic spectrum, a much better gauge of a space rock’s proportions.
Observatories on Earth do have infrared capabilities, but our planet’s atmosphere gets in their way, making it hard for them to offer highly accurate readings of an asteroid’s size.
But the James Webb Space Telescope (JWST), hanging out in space, doesn’t have that problem.
Asteroid 2024 YR4 is the smallest object targeted by JWST to date, and one of the smallest objects to have its size directly measured. Observations were taken using both its NIRCam (Near-Infrared Camera) and MIRI (Mid-Infrared Instrument) to study the thermal properties of the asteroid.
NASA, ESA, CSA, A. RIVKIN (APL), A. PAGAN (STSCI)
This observatory, which sits at a gravitationally stable point about a million miles from Earth, is polymathic. Its sniper-like scope can see in the infrared and allows it to peer at the edge of the observable universe, meaning it can study galaxies that formed not long after the Big Bang. It can even look at the light passing through the atmospheres of distant planets to ascertain their chemical makeups. And its remarkably sharp eye means it can also track the thermal glow of an asteroid long after all ground-based telescopes lose sight of it.
In a fortuitous bit of timing, by the moment 2024 YR4 came along, planetary defenders had recently reasoned that JWST could theoretically be used to track ominous asteroids using its own infrared scope, should the need arise. So after IAWN’s warning went out, operators of JWST ran an analysis: Though the asteroid would vanish from most scopes by late March, this one might be able to see the rock until sometime in May, which would allow researchers to greatly refine their assessment of the asteroid’s orbit and its odds of making Earth impact.
Understanding 2024 YR4’s trajectory was important, but “the size was the main motivator,” says Andy Rivkin, an astronomer at Johns Hopkins University’s Applied Physics Laboratory, who led the proposal to use JWST to observe the asteroid. The hope was that even if the impact odds remained high until 2028, JWST would find that 2024 YR4 was on the smaller side of the 130-to-300-feet size range—meaning it would still be a danger, but a far less catastrophic one.
The JWST proposal was accepted by NASA on February 5. But the earliest it could conduct its observations was early March. And time really wasn’t on Earth’s side.
February 7, 2025
GEMINI SOUTH TELESCOPE, CHILE
“At this point, [2024 YR4] was too faint for the Catalina telescopes,” says Catalina’s Wierzchoś. “In our opinion, this was a big deal.”
So Wierzchoś and his colleagues put in a rare emergency request to commandeer the Gemini Observatory, an internationally funded and run facility featuring two large, eagle-eyed telescopes—one in Chile and one atop Hawaii’s Mauna Kea volcano. Their request was granted, and on February 7, they trained the Chile-based Gemini South telescope onto 2024 YR4.
This composite image was captured by a team of astronomers using the Gemini Multi-Object Spectrograph (GMOS). The hazy dot at the center is asteroid 2024 YR4.
INTERNATIONAL GEMINI OBSERVATORY/NOIRLAB/NSF/AURA/M. ZAMANI
The odds of Earth impact dropped ever so slightly, to 2.2%—a minor, but still welcome, development.
Mid-February 2025
MAGDALENA RIDGE OBSERVATORY, NEW MEXICO
By this point, the roster of 2024 YR4 hunters also included the tiny team operating the Magdalena Ridge Observatory (MRO), which sits atop a tranquil mountain in New Mexico.
“It’s myself and my husband,” says Eileen Ryan, the MRO director. “We’re the only two astronomers running the telescope. We have a daytime technician. It’s kind of a mom-and-pop organization.”
Still, the scope shouldn’t be underestimated. “We can see maybe a cell-phone-size object that’s illuminated at geosynchronous orbit,” Ryan says, referring to objects 22,000 miles away. But its most impressive feature is its mobility. While other observatories have slowly swiveling telescopes, MRO’s scope can move like the wind. “We can track the fastest objects,” she says, with a grin—noting that the telescope was built in part to watch missiles for the US Air Force. Its agility and long-distance vision explain why the Space Force is one of MRO’s major clients: It can be used to spy on satellites and spacecraft anywhere from low Earth orbit right out to the lunar regions. And that meant spying on the super-speedy, super-sneaky 2024 YR4 wasn’t a problem for MRO, whose own observations were vital in refining the asteroid’s impact odds.
Eileen Ryan is the director of the Magdalena Ridge Observatory in New Mexico.
COURTESY PHOTO
Then, in mid-February, MRO and all ground-based observatories came up against an unsolvable problem: The full moon was out, shining so brightly that it blinded any telescope that dared point at the night sky. “During the full moon, the observatories couldn’t observe for a week or so,” says NEOCC’s Fenucci. To most of us, the moon is a beautiful silvery orb. But to astronomers, it’s a hostile actor. “We abhor the moon,” says Denneau.
All any of them could do was wait. Those tracking 2024 YR4 vacillated between being animated and slightly trepidatious. The thought that the asteroid could still stand a decent chance of impacting Earth after it faded from view did weigh a little on their minds.
Nevertheless, Farnocchia maintained his characteristic sangfroid throughout. “I try to stress about the things I can control,” he says. “All we can do is to explain what the situation is, and that we need new data to say more.”
February 18, 2025
CENTER FOR NEAR-EARTH OBJECT STUDIES, CALIFORNIA
As the full moon finally faded into a crescent of light, the world’s largest telescopes sprang back into action for one last shot at glory. “The dark time came again,” says Hainaut, with a smile.
New observations finally began to trickle in, and Sentry, Aegis, and NEODyS readjusted their forecasts. It wasn’t great news: The odds of an Earth impact in 2032 jumped up to 3.1%, officially making 2024 YR4 the most dangerous asteroid ever discovered.
This declaration made headlines across the world—and certainly made the curious public sit up and wonder if they had yet another apocalyptic concern to fret about. But, as ever, the asteroid hunters held fast in their prediction that sooner or later—ideally sooner—more observations would cause those impact odds to drop.
“We kept at it,” says Ryan. But time was running short; they started to “search for out-of-the-box ways to image this asteroid,” says Fenucci.
Planetary defense researchers soon realized that 2024 YR4 wasn’t too far away from NASA’s Lucy spacecraft, a planetary science mission making a series of flybys of various asteroids. If Lucy could be redirected to catch up to 2024 YR4 instead, it would give humanity its best look at the rock, allowing Sentry and company to confirm or dispel our worst fears.
Sadly, NASA ran the numbers, and it proved to be a nonstarter: 2024 YR4 was too speedy and too far for Lucy to pursue.
It was really starting to look as if JWST would be the last, best hope to track 2024 YR4.
A CHANGE OF FATE
February 19, 2025
VERY LARGE TELESCOPE, CHILE and MAGDALENA RIDGE OBSERVATORY, NEW MEXICO
Just one day after 2024 YR made history, the VLT, MRO, and others caught sight of it again, and Sentry, Aegis, and NEODyS voraciously consumed their new data.
The odds of an Earth impact suddenly dropped to 1.5%.
Astronomers didn’t really have time to react to the possibility that this was a good sign—they just kept sending their observations onward.
February 20, 2025
SUBARU TELESCOPE, HAWAII
Yet another observatory had been itching to get into the game for weeks, but it wasn’t until February 20 that Tsuyoshi Terai, an astronomer at Japan’s Subaru Telescope, sitting atop Mauna Kea, finally caught 2024 YR4 shifting between the stars. He added his data to the stream.
And all of a sudden, the asteroid lost its lethal luster. The odds of its hitting Earth were now just 0.3%.
At this point, you might expect that all those tracking it would be in a single control room somewhere, eyes glued to their screens, watching the odds drop before bursting into cheers and rapturous applause. But no—the astronomers who had spent so long observing this asteroid remained scattered across the globe. And instead of erupting into cheers, they exchanged modestly worded emails of congratulations—the digital equivalent of a nod or a handshake.
In late February, data from Tsuyoshi Terai, an astronomer at Japan’s Subaru Telescope, which sits atop Mauna Kea, confirmed that 2024 YR4 was not so lethal after all.
NAOJ
“It was a relief,” says Terai. “I was very pleased that our data contributed to put an end to the risk of 2024 YR4.”
February 24, 2025
INTERNATIONAL ASTEROID WARNING NETWORK
Just a few days later, and thanks to a litany of observations continuing to flood in, IAWN issued the all-clear. This once-ominous asteroid’s odds of inconveniencing our planet were at 0.004%—one in 25,000. Today, the odds of an Earth impact in 2032 are exactly zero.
“It was kinda fun while it lasted,” says Denneau.
Planetary defenders may have had a blast defending the world, but these astronomers still took the cosmic threat deeply seriously—and never once took their eyes off the prize. “In my mind, the observers and orbit teams were the stars of this story,” says Fast, NASA’s acting planetary defense officer.
Farnocchia shrugs off the entire thing. “It was the expected outcome,” he says. “We just didn’t know when that would happen.”
Looking back on it now, though, some 2024 YR4 trackers are allowing themselves to dwell on just how close this asteroid came to being a major danger. “It’s wild to watch it all play out,” says Denneau. “We were weeks away from having to spin up some serious mitigation planning.” But there was no need to work out how the save the world. It turned out that 2024 YR4 was never a threat to begin with—it just took a while to check.
And these experiences in handling a dicey space rock will only serve to make the world a safer place to live. One day, an asteroid very much like 2024 YR4 will be seen heading straight for Earth. And those tasked with tracking it will be officially battle-tested, and better prepared than ever to do what needs to be done.
A POSTSCRIPT
March 27, 2025
JAMES WEBB SPACE TELESCOPE, DEEP SPACE, ONE MILLION MILES FROM EARTH
But the story of 2024 YR4 is not quite over—in fact, if this were a movie, it would have an after-credits scene.
After the Earth-impact odds fell off a cliff, JWST went ahead with its observations in March anyway. It found out that 2024 YR4 was 200 feet across—so large that a direct strike on a city would have proved horrifically lethal. Earth just didn’t have to worry about it anymore.
But the moon might. Thanks in part to JWST, astronomers calculated a 3.8% chance that 2024 YR4 will impact the lunar surface in 2032. Additional JWST observations in May bumped those odds up slightly, to 4.3%, and the orbit can no longer be refined until the asteroid’s next Earth flyby in 2028.
“It may hit the moon!” says Denneau. “Everybody’s still very excited about that.”
A lunar collision would give astronomers a wonderful opportunity not only to study the physics of an asteroid impact, but also to demonstrate to the public just how good they are at precisely predicting the future motions of potentially lethal space rocks. “It’s a thing we can plan for without having to defend the Earth,” says Denneau.
If 2024 YR4 is truly going to smash into the moon, the impact—likely on the side facing Earth—would unleash an explosion equivalent to hundreds of nuclear bombs. An expansive crater would be carved out in the blink of an eye, and a shower of debris would erupt in all directions.
None of this supersonic wreckage would pose any danger to Earth, but it would look spectacular: You’d be able to see the bright flash of the impact from terra firma with the naked eye.
“If that does happen, it’ll be amazing,” says Denneau. It will be a spectacular way to see the saga of 2024 YR4—once a mere speck on his computer screen—come to an explosive end, from a front-row seat.
Robin George Andrews is an award-winning science journalist and doctor of volcanoes based in London. He regularly writes about the Earth, space, and planetary sciences, and is the author of two critically acclaimed books: Super Volcanoes (2021) and How to Kill An Asteroid (2024).
Today’s AI landscape is defined by the ways in which neural networks are unlike human brains. A toddler learns how to communicate effectively with only a thousand calories a day and regular conversation; meanwhile, tech companies are reopening nuclear power plants, polluting marginalized communities, and pirating terabytes of books in order to train and run their LLMs.
But neural networks are, after all, neural—they’re inspired by brains. Despite their vastly different appetites for energy and data, large language models and human brains do share a good deal in common. They’re both made up of millions of subcomponents: biological neurons in the case of the brain, simulated “neurons” in the case of networks. They’re the only two things on Earth that can fluently and flexibly produce language. And scientists barely understand how either of them works.
I can testify to those similarities: I came to journalism, and to AI, by way of six years of neuroscience graduate school. It’s a common view among neuroscientists that building brainlike neural networks is one of the most promising paths for the field, and that attitude has started to spread to psychology. Last week, the prestigious journal Nature published a pair of studies showcasing the use of neural networks for predicting how humans and other animals behave in psychological experiments. Both studies propose that these trained networks could help scientists advance their understanding of the human mind. But predicting a behavior and explaining how it came about are two very different things.
In one of the studies, researchers transformed a large language model into what they refer to as a “foundation model of human cognition.” Out of the box, large language models aren’t great at mimicking human behavior—they behave logically in settings where humans abandon reason, such as casinos. So the researchers fine-tuned Llama 3.1, one of Meta’s open-source LLMs, on data from a range of 160 psychology experiments, which involved tasks like choosing from a set of “slot machines” to get the maximum payout or remembering sequences of letters. They called the resulting model Centaur.
Compared with conventional psychological models, which use simple math equations, Centaur did a far better job of predicting behavior. Accurate predictions of how humans respond in psychology experiments are valuable in and of themselves: For example, scientists could use Centaur to pilot their experiments on a computer before recruiting, and paying, human participants. In their paper, however, the researchers propose that Centaur could be more than just a prediction machine. By interrogating the mechanisms that allow Centaur to effectively replicate human behavior, they argue, scientists could develop new theories about the inner workings of the mind.
But some psychologists doubt whether Centaur can tell us much about the mind at all. Sure, it’s better than conventional psychological models at predicting how humans behave—but it also has a billion times more parameters. And just because a model behaves like a human on the outside doesn’t mean that it functions like one on the inside. Olivia Guest, an assistant professor of computational cognitive science at Radboud University in the Netherlands, compares Centaur to a calculator, which can effectively predict the response a math whiz will give when asked to add two numbers. “I don’t know what you would learn about human addition by studying a calculator,” she says.
Even if Centaur does capture something important about human psychology, scientists may struggle to extract any insight from the model’s millions of neurons. Though AI researchers are working hard to figure out how large language models work, they’ve barely managed to crack open the black box. Understanding an enormous neural-network model of the human mind may not prove much easier than understanding the thing itself.
One alternative approach is to go small. The second of the two Nature studies focuses on minuscule neural networks—some containing only a single neuron—that nevertheless can predict behavior in mice, rats, monkeys, and even humans. Because the networks are so small, it’s possible to track the activity of each individual neuron and use that data to figure out how the network is producing its behavioral predictions. And while there’s no guarantee that these models function like the brains they were trained to mimic, they can, at the very least, generate testable hypotheses about human and animal cognition.
There’s a cost to comprehensibility. Unlike Centaur, which was trained to mimic human behavior in dozens of different tasks, each tiny network can only predict behavior in one specific task. One network, for example, is specialized for making predictions about how people choose among different slot machines. “If the behavior is really complex, you need a large network,” says Marcelo Mattar, an assistant professor of psychology and neural science at New York University who led the tiny-network study and also contributed to Centaur. “The compromise, of course, is that now understanding it is very, very difficult.”
This trade-off between prediction and understanding is a key feature of neural-network-driven science. (I also happen to be writing a book about it.) Studies like Mattar’s are making some progress toward closing that gap—as tiny as his networks are, they can predict behavior more accurately than traditional psychological models. So is the research into LLM interpretability happening at places like Anthropic. For now, however, our understanding of complex systems—from humans to climate systems to proteins—is lagging farther and farther behind our ability to make predictions about them.
This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here.
Fusion energy holds the potential to shift a geopolitical landscape that is currently configured around fossil fuels. Harnessing fusion will deliver the energy resilience, security, and abundance needed for all modern industrial and service sectors. But these benefits will be controlled by the nation that leads in both developing the complex supply chains required and building fusion power plants at scales large enough to drive down economic costs.
The US and other Western countries will have to build strong supply chains across a range of technologies in addition to creating the fundamental technology behind practical fusion power plants. Investing in supply chains and scaling up complex production processes has increasingly been a strength of China’s and a weakness of the West, resulting in the migration of many critical industries from the West to China. With fusion, we run the risk that history will repeat itself. But it does not have to go that way.
The US and Europe were the dominant public funders of fusion energy research and are home to many of the world’s pioneering private fusion efforts. The West has consequently developed many of the basic technologies that will make fusion power work. But in the past five years China’s support of fusion energy has surged, threatening to allow the country to dominate the industry.
The industrial base available to support China’s nascent fusion energy industry could enable it to climb the learning curve much faster and more effectively than the West. Commercialization requires know-how, capabilities, and complementary assets, including supply chains and workforces in adjacent industries. And especially in comparison with China, the US and Europe have significantly under-supported the industrial assets needed for a fusion industry, such as thin-film processing and power electronics.
To compete, the US, allies, and partners must invest more heavily not only in fusion itself—which is already happening—but also in those adjacent technologies that are critical to the fusion industrial base.
China’s trajectory to dominating fusion and the West’s potential route to competing can be understood by looking at today’s most promising scientific and engineering pathway to achieve grid-relevant fusion energy. That pathway relies on the tokamak, a technology that uses a magnetic field to confine ionized gas—called plasma—and ultimately fuse nuclei. This process releases energy that is converted from heat to electricity. Tokamaks consist of several critical systems, including plasma confinement and heating, fuel production and processing, blankets and heat flux management, and power conversion.
A close look at the adjacent industries needed to build these critical systems clearly shows China’s advantage while also providing a glimpse into the challenges of building a fusion industrial base in the US or Europe. China has leadership in three of these six key industries, and the West is at risk of losing leadership in two more. China’s industrial might in thin-film processing, large metal-alloy structures, and power electronics provides a strong foundation to establish the upstream supply chain for fusion.
The importance of thin-film processing is evident in the plasma confinement system. Tokamaks use strong electromagnets to keep the fusion plasma in place, and the magnetic coils must be made from superconducting materials. Rare-earth barium copper oxide (REBCO) superconductors are the highest-performing materials available in sufficient quantity to be viable for use in fusion.
The REBCO industry, which relies on thin-film processing technologies, currently has low production volumes spanning globally distributed manufacturers. However, as the fusion industry grows, the manufacturing base for REBCO will likely consolidate among the industry players who are able to rapidly take advantage of economies of scale. China is today’s world leader in thin-film, high-volume manufacturing for solar panels and flat-panel displays, with the associated expert workforce, tooling sector, infrastructure, and upstream materials supply chain. Without significant attention and investment on the part of the West, China is well positioned to dominate REBCO thin-film processing for fusion magnets.
The electromagnets in a full-scale tokamak are as tall as a three-story building. Structures made using strong metal alloys are needed to hold these electromagnets around the large vacuum vessel that physically contains the magnetically confined plasma. Similar large-scale, complex metal structures are required for shipbuilding, aerospace, oil and gas infrastructure, and turbines. But fusion plants will require new versions of the alloys that are radiation-tolerant, able to withstand cryogenic temperatures, and corrosion-resistant. China’s manufacturing capacity and its metallurgical research efforts position it well to outcompete other global suppliers in making the necessary specialty metal alloys and machining them into the complex structures needed for fusion.
A tokamak also requires large-scale power electronics. Here again China dominates. Similar systems are found in the high-speed rail (HSR) industry, renewable microgrids, and arc furnaces. As of 2024, China had deployed over 48,000 kilometers of HSR. That is three times the length of Europe’s HSR network and 55 times as long as the Acela network in the US, which is slower than HSR. While other nations have a presence, China’s expertise is more recent and is being applied on a larger scale.
But this is not the end of the story. The West still has an opportunity to lead the other three adjacent industries important to the fusion supply chain: cryo-plants, fuel processing, and blankets.
The electromagnets in an operational tokamak need to be kept at cryogenic temperatures of around 20 Kelvin to remain superconducting. This requires large-scale, multi-megawatt cryogenic cooling plants. Here, the country best set up to lead the industry is less clear. The two major global suppliers of cryo-plants are Europe-based Linde Engineering and Air Liquide Engineering; the US has Air Products and Chemicals and Chart Industries. But they are not alone: China’s domestic champions in the cryogenic sector include Hangyang Group, SASPG, Kaifeng Air Separation, and SOPC. Each of these regions already has an industrial base that could scale up to meet the demands of fusion.
Fuel production for fusion is a nascent part of the industrial base requiring processing technologies for light-isotope gases—hydrogen, deuterium, and tritium. Some processing of light-isotope gases is already done at small scale in medicine, hydrogen weapons production, and scientific research in the US, Europe, and China. But the scale needed for the fusion industry does not exist in today’s industrial base, presenting a major opportunity to develop the needed capabilities.
Similarly, blankets and heat flux management are an opportunity for the West. The blanket is the medium used to absorb energy from the fusion reaction and to breed tritium. Commercial-scale blankets will require entirely novel technology. To date, no adjacent industries have relevant commercial expertise in liquid lithium, lead-lithium eutectic, or fusion-specific molten salts that are required for blanket technology. Some overlapping blanket technologies are in early-stage development by the nuclear fission industry. As the largest producer of beryllium in the world, the US has an opportunity to capture leadership because that element is a key material in leading fusion blanket concepts. But the use of beryllium must be coupled with technology development programs for the other specialty blanket components.
These six industries will prove critical to scaling fusion energy. In some, such as thin-film processing and large metal-alloy structures, China already has a sizable advantage. Crucially, China recognizes the importance of these adjacent industries and is actively harnessing them in its fusion efforts. For example, China launched a fusion consortium that consists of industrial giants spanning the steel, machine tooling, electric grid, power generation, and aerospace sectors. It will be extremely difficult for the West to catch up in these areas, but policymakers and business leaders must pay attention and try to create robust alternative supply chains.
As the industrial area of greatest strength, cryo-plants could continue to be an opportunity for leadership in the West. Bolstering Western cryo-plant production by creating demand for natural-gas liquefaction will be a major boon to the future cryo-plant supply chain that will support fusion energy.
The US and European countries also have an opportunity to lead in the emerging industrial areas of fuel processing and blanket technologies. Doing so will require policymakers to work with companies to ensure that public and private funding is allocated to these critical emerging supply chains. Governments may well need to serve as early customers and provide debt financing for significant capital investment. Governments can also do better to incentivize private capital and equity financing—for example, through favorable capital-gains taxation. In lagging areas of thin-film and alloy production, the US and Europe will likely need partners, such as South Korea and Japan, that have the industrial bases to compete globally with China.
The need to connect and capitalize multiple industries and supply chains will require long-term thinking and clear leadership. A focus on the demand side of these complementary industries is essential. Fusion is a decade away from maturation, so its supplier base must be derisked and made profitable in the near term by focusing on other primary demand markets that contribute to our economic vitality. To name a few, policymakers can support modernization of the grid to bolster domestic demand for power electronics and domestic semiconductor manufacturing to support thin-film processing.
The West must also focus on the demand for energy production itself. As the world’s largest energy consumer, China will leverage demand from its massive domestic market to climb the learning curve and bolster national champions. This is a strategy that China has wielded with tremendous success to dominate global manufacturing, most recently in the electric-vehicle industry. Taken together, supply- and demand-side investment have been a winning strategy for China.
The competition to lead the future of fusion energy is here. Now is the moment for the US and its Western allies to start investing in the foundational innovation ecosystem needed for a vibrant and resilient industrial base to support it.
Daniel F. Brunner is a co-founder of Commonwealth Fusion Systems and a Partner at Future Tech Partners.
Edlyn V. Levine is the co-founder of a stealth-mode technology start up and an affiliate of the MIT Sloan School of Management.
Fiona E. Murray is a professor of entrepreneurship at the MIT School of Management and Vice Chair of the NATO Innovation Fund.
Rory Burke is a graduate of MIT Sloan and a former summer scholar with ARPA-E.
The findings a decade ago were, well, shocking. Mainstream economists had long argued that free trade was overall a good thing; though there might be some winners and losers, it would generally bring lower prices and widespread prosperity. Then, in 2013, a trio of academic researchers showed convincing evidence that increased trade with China beginning in the early 2000s and the resulting flood of cheap imports had been an unmitigated disaster for many US communities, destroying their manufacturing lifeblood.
The results of what in 2016 they called the “China shock” were gut-wrenching: the loss of 1 million US manufacturing jobs and 2.4 million jobs in total by 2011. Worse, these losses were heavily concentrated in what the economists called “trade-exposed” towns and cities (think furniture makers in North Carolina).
If in retrospect all that seems obvious, it’s only because the research by David Autor, an MIT labor economist, and his colleagues has become an accepted, albeit often distorted, political narrative these days: China destroyed all our manufacturing jobs! Though the nuances of the research are often ignored, the results help explain at least some of today’s political unrest. It’s reflected in rising calls for US protectionism, President Trump’s broad tariffs on imported goods, and nostalgia for the lost days of domestic manufacturing glory.
The impacts of the original China shock still scar much of the country. But Autor is now concerned about what he considers a far more urgent problem—what some are calling China shock 2.0. The US, he warns, is in danger of losing the next great manufacturing battle, this time over advanced technologies to make cars and planes as well as those enabling AI, quantum computing, and fusion energy.
Recently, I asked Autor about the lingering impacts of the China shock and the lessons it holds for today’s manufacturing challenges.
How are the impacts of the China shock still playing out?
I have a recent paper looking at 20 years of data, from 2000 to 2019. We tried to ask two related questions. One, if you looked at the places that were most exposed, how have they adjusted? And then if you look to the people who are most exposed, how have they adjusted? And how do those two things relate to one anothe
It turns out you get two very different answers. If you look at places that were most exposed, they have been substantially transformed. Manufacturing, once it starts going down, never comes back. But after 2010, these trade-impacted local labor markets staged something of an employment recovery, such that employment has grown faster after 2010 in trade-exposed places than non-trade-exposed places because a lot of people have come in. But these are jobs mostly in low-wage sectors. They’re in K–12 education and non-traded health services. They’re in warehousing and logistics. They’re in hospitality and lodging and recreation, and so they’re lower-wage, non-manufacturing jobs. And they’re done by a really different set of people.
The growth in employment is among women, among native-born Hispanics, among foreign-born adults and a lot of young people. The recovery is staged by a very different group from the white and black men, but especially white men, who were most represented in manufacturing. They have not really participated in this renaissance.
Employment is growing, but are these areas prospering?
They have a lower wage structure: fewer high-wage jobs, more low-wage jobs. So they’re not, if your definition of prospering is rapidly rising incomes. But there’s a lot of employment growth. They’re not like ghost towns. But then if you look at the people who were most concentrated in manufacturing—mostly white, non-college, native-born men—they have not prospered. Most of them have not transitioned from manufacturing to non-manufacturing.
One of the great surprises is everyone had believed that people would pull up stakes and move on. In fact, we find the opposite. People in the most adversely exposed places become less likely to leave. They have become less mobile. The presumption was that they would just relocate to find higher ground. And that is not at all what occurred.
What happened to the total number of manufacturing jobs?
There’s been no rebound. Once they go, they just keep going. If there is going to be new manufacturing, it won’t be in the sectors that were lost to China. Those were basically labor-intensive jobs, the kind of low-tech sectors that we will not be getting back. You know—commodity furniture and assembly of things, shoes, construction material. The US wasn’t going to keep them forever, and once they’re gone, it’s very unlikely to get them back.
I know you’ve written about this, but it’s not hard to draw a connection between the dynamics you’re describing—white-male manufacturing jobs going away and new jobs going to immigrants—and today’s political turmoil.
How big a factor would you say it is in today’s political unrest?
I don’t want to say it’s the factor. The China trade shock was a catalyst, but there were lots of other things that were happening. It would be a vast oversimplification to say that it was the sole cause.
But most people don’t work in manufacturing anymore. Aren’t these impacts that you’re talking about, including the political unrest, disproportionate to the actual number of jobs lost?
These are jobs in places where manufacturing is the anchor activity. Manufacturing is very unevenly distributed. It’s not like grocery stores and hospitals that you find in every county. The impact of the China trade shock on these places was like dropping an economic bomb in the middle of downtown. If the China trade shock cost us a few million jobs, and these were all—you know—people in groceries and retail and gas stations, in hospitality and in trucking, you wouldn’t really notice it that much. We lost lots of clerical workers over the last couple of decades. Nobody talks about a clerical shock. Why not? Well, there was never a clerical capital of America. Clerical workers are everywhere. If they decline, it doesn’t wipe out the entire basis of a place.
So it goes beyond the jobs. These places lost their identity.
Maybe. But it’s also the jobs. Manufacturing offered relatively high pay to non-college workers, especially non-college men. It was an anchor of a way of life.
And we’re still seeing the damage.
Yeah, absolutely. It’s been 20 years. What’s amazing is the degree of stasis among the people who are most exposed—not the places, but the people. Though it’s been 20 years, we’re still feeling the pain and the political impacts from this transition.
Clearly, it has now entered the national psyche. Even if it weren’t true, everyone now believes it to have been a really big deal, and they’re responding to it. It continues to drive policy, political resentments, maybe even out of proportion to its economic significance. It certainly has become mythological.
What worries you now?
We’re in the midst of a totally different competition with China now that’s much, much more important. Now we’re not talking about commodity furniture and tube socks. We’re talking about semiconductors and drones and aviation, electric vehicles, shipping, fusion power, quantum, AI, robotics. These are the sectors where the US still maintains competitiveness, but they’re extremely threatened. China’s capacity for high-tech, low-cost, incredibly fast, innovative manufacturing is just unbelievable. And the Trump administration is basically fighting the war of 20 years ago. The loss of those jobs, you know, was devastating to those places. It was not devastating to the US economy as a whole. If we lose Boeing, GM, and Apple and Intel—and that’s quite possible—then that will be economically devastating.
I think some people are calling it China shock 2.0.
Yeah. And it’s well underway.
When we think about advanced manufacturing and why it’s important, it’s not so much about the number of jobs anymore, is it? Is it more about coming up with the next technologies?
It does create good jobs, but it’s about economic leadership. It’s about innovation. It’s about political leadership, and even standard setting for how the rest of the world works.
Should we just accept that manufacturing as a big source of jobs is in the past and move on?
No. It’s still 12 million jobs, right? Instead of the fantasy that we’re going to go back to 18 million or whatever—we had, what, 17.7 million manufacturing jobs in 1999—we should be worried about the fact that we’re going to end up at 6 million, that we’re going to lose 50% in the next decade. And that’s quite possible. And the Trump administration is doing a lot to help that process of loss along.
We have a labor market of over 160 million people, so it’s like 8% of employment. It’s not zero. So you should not think of it as too small to worry about it. It’s a lot of people; it’s a lot of jobs. But more important, it’s a lot of what has helped this country be a leader. So much innovation happens here, and so many of the things in which other countries are now innovating started here. It’s always been the case that the US tends to innovate in sectors and then lose them after a while and move on to the next thing. But at this point, it’s not clear that we’ll be in the frontier of a lot of these sectors for much longer.
So we want to revive manufacturing, but the right kind—advanced manufacturing?
The notion that we should be assembling iPhones in the United States, which Trump wants, is insane. Nobody wants to do that work. It’s horrible, tedious work. It pays very, very little. And if we actually did it here, it would make the iPhones 20% more expensive or more. Apple may very well decide to pay a 25% tariff rather than make the phones here. If Foxconn started doing iPhone assembly here, people would not be lining up for that job.
But at the same time, we do need new people coming into manufacturing.
But not that manufacturing. Not tedious, mind-numbing, eyestrain-inducing assembly.
We need them to do high-tech work. Manufacturing is a skilled activity. We need to build airplanes better. That takes a ton of expertise. Assembling iPhones does not.
What are your top priorities to head off China shock 2.0?
I would choose sectors that are important, and I would invest in them. I don’t think that tariffs are never justified, or industrial policies are never justified. I just don’t think protecting phone assembly is smart industrial policy. We really need to improve our ability to make semiconductors. I think that’s important. We need to remain competitive in the automobile sector—that’s important. We need to improve aviation and drones. That’s important. We need to invest in fusion power. That’s important. We need to adopt robotics at scale and improve in that sector. That’s important. I could come up with 15 things where I think public money is justified, and I would be willing to tolerate protections for those sectors.
What are the lasting lessons of the China shock and the opening up of global trade in the 2000s?
We did it too fast. We didn’t do enough to support people, and we pretended it wasn’t going on.
When we started the China shock research back around 2011, we really didn’t know what we’d find, and so we were as surprised as anyone. But the work has changed our own way of thinking and, I think, has been constructive—not because it has caused everyone to do the right thing, but it at least caused people to start asking the right questions.
What do the findings tell us about China shock 2.0?
I think the US is handling that challenge badly. The problem is much more serious this time around. The truth is, we have a sense of what the threats are. And yet we’re not seemingly responding in a very constructive way. Although we now know how seriously we should take this, the problem is that it doesn’t seem to be generating very serious policy responses. We’re generating a lot of policy responses—they’re just not serious ones.
