EXECUTIVE SUMMARY GLP-1 weight-loss drugs are one of the biggest biotech breakthroughs of the last decade. But to some clinical psychologists, they can represent something more sinister—a potential trigger for eating disorders. A new report suggests that the global incidence of anorexia nervosa and bulimia nervosa shot up between 1990 and 2023—by a whopping 71%. By 2023, there were an estimated 18.1 million people living with one of these two disorders. What’s worse is that the number is likely to be much higher than that once we factor in other eating disorders. And, disturbingly, those in the field are already starting to see GLP-1 weight-loss drugs exacerbating the issue. Cheri Levinson is a clinical psychologist at the University of Louisville in Kentucky who has specialized in eating disorders for almost 20 years. She says she’s now seeing more patients with more severe symptoms than at any other time in her career, and that many of them are misusing GLP-1s. “This is the worst I’ve seen things,” she says.
Eating disorders are mental health conditions in which people’s eating behaviors impact their health. They include anorexia nervosa (defined by a fear of weight gain), bulimia nervosa (in which people binge and purge), binge eating disorder, and others. It’s thought that some people have a genetic predisposition to certain eating disorders, but that some external factor is what triggers their onset. That might be stress, or social pressure to be skinny. Often, simply restricting food intake can trigger an eating disorder, says Levinson.
Eating disorders—along with the pressure to look thin—were already prevalent in the 1990s (the heroin chic era springs to mind). To find out how things might have changed since then, Ruhai Bai at the Children’s Hospital of Nanjing Medical University in China and his colleagues pulled together data from studies and surveys conducted on populations around the world. The team used statistical models to fill in any data gaps. That research allowed the researchers to estimate the annual incidence of two eating disorders—anorexia nervosa and bulimia nervosa—for each of 204 countries and territories between 1990 and 2023. They also estimated the impact those disorders had on global health by calculating their disability-adjusted life years (one DALY represents a year of healthy life lost). The findings, published in JAMA Psychiatry on Wednesday, are unsettling. Between 1990 and 2023, there was a 71% increase in the incidence of those two eating disorders, the team found. Those conditions had a huge impact on global health, too. In 2023, they “cost the world the equivalent of about 3.83 million years of healthy life,” says Bai. Bai and his colleagues expect these figures to continue to rise. In 2023, an estimated 18.1 million people had either anorexia nervosa or bulimia nervosa; by 2050, the team predict, the two conditions could affect 21.8 million people. It gets worse. Those two conditions represent less than half of all eating disorders, says Levinson, who was not involved in Bai’s study. The true impact of these disorders is “probably three times the rate that they’re reporting,” she says. Bai and his colleagues didn’t explore the reasons for this rise (although Bai points out that it can’t be entirely explained by population growth). So how did we get here? Social media may have played a role. “Social media comparisons, influencers, challenges, diet trends can all be precipitating … events for an eating disorder,” says Cynthia Bulik, a clinical psychologist at the University of North Carolina, who was not involved in the new study. Bulik has been studying and treating eating disorders since the early 1980s. “Between 1990 and 2023 there has clearly been an upsurge of those types of environmental exposures,” she says. Many social media users are exposed to filtered and modified images of skinny influencers who may have undergone cosmetic surgery or extreme dieting and who are promoting “thinspiration” or “fitspiration” content. But users with eating disorders appear to be more likely to engage with this content—and to be fed more of it as a result.

When researchers at the University of Melbourne investigated this phenomenon a couple of years ago, they found that the algorithms of TikTok users with eating disorders delivered 4343% more “toxic eating disorder videos” than those of users who didn’t have the disorders. They also got more content focused on dieting or exercising. Many were fed over 1000 problematic videos per month. Bulik describes social media as “quicksand” for people with, or at risk of, eating disorders. The covid-19 pandemic—and the social isolation that accompanied lockdowns—also appears to have contributed to the rise in eating disorders. Research conducted in 2021 found that the number of people hospitalized with eating disorders in the US doubled between 2018 and 2020. Researchers don’t yet have good data on how trends in eating disorders might have progressed since 2023. But anecdotally, things seem to be getting even worse. “There are more eating disorders, more severe eating disorders… and we’re seeing earlier ages of onset,” says Levinson. “We’re seeing eight- and nine-year-olds with eating disorders.” “From what I’m hearing and seeing, I am not going to be surprised if that number isn’t at least doubling again,” she adds. Levinson particularly worries about the impact of GLP-1 weight-loss drugs, which have exploded in popularity in recent years. A Gallup survey conducted earlier this year suggested that 15% of Americans had used one of these drugs for weight loss. And they’re becoming more accessible—drug companies are working on new formulations that are both easier to take and more effective, and many people can buy versions of these drugs from compound pharmacies or dubious online sources. In a small study published earlier this year, Levinson and her colleagues found that 30% of people with eating disorders had used GLP-1 drugs. The drugs might benefit some of those people. But 10% appeared to have misused the drugs. And Levinson thinks that figure is also an underestimate. Levinson says she and her colleagues have seen children develop eating disorders after being prescribed GLP-1s for medical reasons. They’ve seen others—even children as young as 11—buy versions of these drugs over TikTok. “We’re very concerned that GLPs will lead to even faster and more extreme weight loss in people with anorexia nervosa, and that patients will come to us in even more advanced stages of illness that become difficult to reverse,” says Bulik.
Both Bulik and Levinson are continuing to study the impacts of these drugs in eating disorders. In the meantime, both urge medical professionals to consider the risks GLP-1s might pose to people with those disorders. Bulik contributed to a set of research and clinical guidelines, published earlier this month, that recommends doctors screen for eating disorders before prescribing the drugs, for example. The guidelines also recommend more patient education and more robust monitoring of people who take GLP-1s. If you think you or someone you know might have an eating disorder, you can find information and support at the National Eating Disorders Association. 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.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Opinion: We urgently need to monitor AI agents’ actions in real time —Madhulika Srikumar and Claire Leibowicz from the Partnership on AI, a nonprofit advancing best practices in AI governance. AI agents are taking unauthorized actions, from hacking websites to accessing private government files. In many cases, they’ve eluded the very researchers and safety teams tasked with monitoring them.  As agents roll out across critical sectors, from finance to government to medicine, we must be able to monitor them in real time to prevent potentially devastating consequences. And while safety depends on monitoring, monitoring requires observability.
​​In other words, we need to know what the agent did and why it did it. Because right now, we’re effectively flying blind when deploying these tools.  Here’s what needs to change.
Our climate models have a big, dangerous blind spot Climate change itself increasingly fuels climate change. Rising greenhouse gas concentrations are driving up global temperatures, creating dangerous feedback loops that aren’t fully represented in our climate models—or our emissions-reduction goals. Scientists have long warned that warming would propel further warming. Now, a growing number of observations and studies are indicating how powerful these cascading effects could be. By extension, they’re also showing how much faster the planet may warm this century. Read more about the dangerous feedback loops accelerating global warming. —James Temple This article is from The Spark, our weekly climate tech newsletter. Sign up to receive it in your inbox every Wednesday. Are we there yet? Solid-state batteries  Lithium-ion batteries power everything from personal electronics to grid-scale energy storage, but their liquid electrolytes have drawbacks, not least a troubling tendency to catch fire. Enter the solid-state battery, which swaps liquid for a solid ceramic or polymer. The technology promises faster charging, longer life, greater energy density and improved safety, and a new generation of companies aims to bring it to market. If their bets pan out, the tech could ease the shift away from fossil fuels. Find out how close solid-state batteries are to delivering on their promise.

—Alexander C. Kaufman This story is part of “Are we there yet?”, our series that takes stock of six technologies that have long promised to change the world—and scores them on a progress meter. We’re exploring each one in The Download, so stay tuned, and subscribe to read the rest of the stories in our new Impossible issue. The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 Hackers exposed 41,000 Homeland Security workersIncluding undercover ICE agents. (Wired $)+ Much of the data relates to Customs and Border Protection. (TechCrunch)+ It also linked the agencies to specific surveillance tools. (Reuters $)+ Big Tech is pushing back against ICE’s demands for user data. (WSJ $)+ Activists want ChatGPT users to quit over ICE ties. (MIT Technology Review) 2 Putin says Russia is coordinating with “genius” Musk on satellitesHe said the cooperation maintains order in orbit. (Bloomberg $)+ ​​Russia is developing its own version of Starlink. (Guardian)+ Moscow added Apple and Microsoft to a surveillance registry. (Quartz)+ Here’s all the stuff we’ve put up in space. (MIT Technology Review) 3 Apple just announced its first-ever MacBook with a touchscreenIt joins a blitz of product launches under new CEO John Ternus. (CNN)+ Steve Jobs and Tim Cook had both dismissed the idea. (NYT $)+ Executives say touchscreen advances mean it’s now worthwhile. (Wired $) 4 Bloomberg and trading firms are paying to get Trump’s posts fasterThey’ve signed up for Truth Social’s pay-for-access plan. (Reuters $)
5 As US ties fray, Europe is planning its first subsea cable to GreenlandIt will bypass the US to link Canada, Japan, and South Korea. (Ars Technica)+ The EU’s AI champion says Europe can rival US tech giants. (FT $) 6 Palantir’s defense chief says AI could make China target TaiwanHe says Beijing may act to stop the US pulling ahead in AI. (Bloomberg $)+ Taiwan’s “silicon shield” could be weakening. (MIT Technology Review)
7 Musk’s xAI has offered to buy homes if residents drop a noise lawsuitThe suit targets noise from powering its Mississippi data centres. (Guardian) 8 The world’s first spinal cord transplant has sparked controversyExperts say there’s little evidence the operation will work. (New Scientist $) 9 China’s tech giants are fighting for attention with AI microdramasTencent and ByteDance are battling over the booming market. (SCMP)+ Chinese microdramas are AI content machines. (MIT Technology Review) 10 Astronomers have spotted a possible new dwarf planet beyond PlutoThe icy world would be the solar system’s sixth dwarf planet. (Gizmodo) Quote of the day “They have our information, so it’s only fair that we have theirs.”  —A hacking group explains why it leaked personal information from the US Department of Homeland Security, NPR reports.
One more thing How the Pentagon is adapting to China’s technological rise It’s been just over a year since Kathleen Hicks stepped down as US deputy secretary of defense. As the highest-ranking woman in Pentagon history, Hicks shaped US military posture through an era defined by renewed competition between powerful countries and a scramble to modernize defense technology. In this conversation with MIT Technology Review, Hicks reflects on how the Pentagon is adapting—or failing to adapt—to a new era of geopolitical competition. She discusses China’s technological rise, the future of AI in warfare, and her signature initiative: Replicator. 

EXECUTIVE SUMMARY Climate change itself is increasingly fueling climate change. Rising concentrations of greenhouse gases are driving up global temperatures, setting off dangerous feedback loops that we have not fully accounted for in our climate models—or our emissions-reduction goals. Scientists have long warned that warming would propel further warming in various ways, but a growing number of observations and studies are at last providing a sharper sense of just how powerful these cascading effects could be. And, by extension, they’re showing how much faster the planet may warm in this century. A study in Environmental Research Letters in September, synthesizing the results of previous papers, found that the additional carbon dioxide and methane released from thawing permafrost, intensifying wildfires, hotter lakes and reservoirs, and wetter, warmer wetlands alone could drive global average temperatures up 0.2 °C to 0.4 °C this century. That would come on top of the 2.6°C of warming above preindustrial levels that modelers are already expecting through 2100. The additional heat could subject many of the world’s cities to monthslong heat waves and expose hundreds of millions of people to flooding.  None of the models that inform the major UN climate reports, which assess how much the planet will likely warm under various emissions scenarios, account for all these feedback effects. Many of them don’t incorporate any. That leaves a huge blind spot in our understanding of how much the world could actually warm in the coming decades—and how rapidly nations will need to slash emissions to stay within critical boundaries.
A separate analysis published in One Earth in March, which examined what the feedback problem means for our international climate goals, concluded that the world may be much further off track than is widely believed. It found that just three regional sources of warming-driven emissions—thawing permafrost, increasing fires in boreal forests, and warming tropical wetlands—may shorten by as much as 25% the time before the planet consistently breaches the 1.5 °C and 2 °C warming thresholds. Our climate models and the UN’s reports clearly need to be updated to better represent these effects. Nations must also step up their efforts to reduce emissions to reflect this improving understanding of the dangers we face, says Sam Abernethy, the lead author of the paper and a climate scientist at Spark Climate Solutions, a San Francisco–based nonprofit research group. 
At least some of this is starting to happen. Spark kick-started an initiative, called the Warming-Induced Greenhouse Gas Emissions Model Intercomparison Project, through which more than 20 international teams are running the same set of analyses on different models to develop greater understanding of feedback effects’ potential power. Researchers involved hope that the UN will incorporate those results into its next major climate assessment, due out in 2029. That report will update estimates of the remaining emissions that can be released before the world crosses key temperature thresholds. But what is already clear is that the planet is rapidly warming and that nations haven’t done nearly enough to cut fossil-fuel pollution. And even if we did manage to halt it, warming-driven emissions could continue.  So a number of the research groups involved in the recent studies have started to explore what nations could do to get feedback emissions under control. Scientists at the Woodwell Climate Research Center in Massachusetts are, for example, investigating the possibility of preventing fires or rapidly detecting them and putting them out when they threaten to burn the vegetation and soil that help insulate particularly carbon-rich wedges of permafrost (a subject I explored at length in a recent feature story). In addition, the Environmental Defense Fund’s Climate Innovation Initiative has backed lab research on the potential of adding sulfate, iron, or other compounds to wetlands to counteract fast-rising methane emissions.  Several startups are also exploring possibilities. They’re looking into suppressing the lightning strikes that are igniting more and more fires in high latitudes, placing synthetic mats above permafrost to prevent it from melting, and deploying autonomous drones to quickly suppress potentially high-emissions fires before they burn out of control. Investigating all these ideas is worthwhile. But researchers are only in the earliest stages of figuring out how big a dent in emissions any such efforts could achieve across sprawling global ecosystems undergoing massive transformations.  The surest path to reducing the effect of warming-driven warming remains the one we’ve known all along: breaking society’s addiction to fossil fuels.  This article is from The Spark, MIT Technology Review’s weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here. 

This summer, hundreds of OpenAI agents tried to cheat on a cybersecurity test. They broke out of their test environment and successfully hacked Hugging Face, a site that they believed held clues about how they’d be graded. Around the same time, another OpenAI agent accessed private files from an Australian government website. It’s a problem that’s by no means confined to OpenAI. In the months since, Anthropic, Google, and Meta have also disclosed incidents in which their agents have gone off track. And worryingly, in many cases the agents eluded the very researchers and safety teams tasked with monitoring them.  AI agents are taking unauthorized actions, and the companies building and running them are mostly relying on luck and hindsight to mitigate any harmful impacts. They must do more to ensure safety. That’s why EU regulators and US agencies are rightfully calling for agent developers and deployers to monitor agents in real time. But these rules assume that companies can see what their agents are doing and why—when in reality, they lack sufficient visibility.  Imagine a bank hiring a thousand new employees who can access customer accounts and sensitive information, but the employees’ ID badges change every day, it’s not clear whether a manager ever reviewed or authorized their work, and nobody logs the changes employees make to approval processes. That lack of visibility is where we are with agents today. As agents roll out across critical sectors, from finance to government to medicine, we must be able to monitor them in real time to prevent wide-ranging and potentially devastating consequences.  And while safety depends on monitoring, monitoring requires observability: the capacity to capture what the system did, process it into signals, and present those to humans or automated oversight systems. In other words, we need to know what the agent did and why it did it. And as our recent research shows, current agent frameworks—the software that runs agents— do not reliably generate the crucial observability records we need to ensure that agents are deployed safely.
Agents present new kinds of problems that generative AI alone does not. They inherit the unpredictability of the LLMs they are built on, but where a generative AI system produces a single output for a human to act on, an agent performs longer tasks—it plans, picks tools, spawns sub-agents, and takes actions. So an agent’s course can take shape in a multitude of different ways, and no tests before launch can fully anticipate the risks. This makes observing the agents while they work even more challenging, and even more important. With colleagues from academia, civil society, and industry, we studied the four most widely used agent frameworks to see what signals they actually emit. We found major gaps that made it hard to achieve key safety goals, like noticing policy violations, unauthorized actions, and even human oversight failures. Key observability requirements were missing—like Agent IDs that persist over time, and records about which humans approved agent tasks. This made it hard to tell which agents did what or assess whether they were taking unauthorized actions. 
The good news is that these gaps, once identified, can be closed. Doing so starts with capturing two types of records: those that reveal what happened in an agent’s run and how the agent made its decisions. The “what happened” records identify the agent (while preserving privacy) and log how humans intervened and whether agent permissions or “memory” shifted, while the “decision” records show an agent’s reasoning and levels of confidence in the actions it’s taken. Take, for example, agents working to process a bank’s mortgage loans. The system examines an application and recommends denying it, and then a human loan officer signs off on the decision. In a later audit, or perhaps even in real time, a bank or regulator needs to be able to reconstruct which agent made the call, which officer approved it, and how long they took. And the bank must also analyze records of how the agent arrived at its refusal decision in order to ensure it did not use off-limits information, like details about the neighborhood the applicant is from. Recording “what happened” and “how the agent decided” require different approaches, but both are feasible. The current lack of “what happened” records is mostly a software plumbing problem that can be addressed through OpenTelemetry, an open observability standard maintained by engineers across the industry and by the agent frameworks adopting it. OpenTelemetry has launched a working group drafting shared formats for tracing agents, but they have yet to be finalized. The group should urgently incorporate standards for maintaining persistent agent identity and for recording human interventions and changes to agents’ permission or memory. And it should be done before these systems can scale, all the while attending to the privacy implications of agent ID that links to real people. Preserving “how agents decided” records is more challenging. The issue raises open research questions for model providers and researchers: Does the step-by-step reasoning a model displays, known as chain of thought, reflect how the model actually made a decision? And do we have reliable ways to measure how confident the agent is in that decision? These questions need answers before anyone sets standards for how such observability records are used. No single organization or sector can close the observability gaps on their own, but together, they can support the infrastructure required to prevent dangerous incidents and support safety.  AI developers building agent frameworks, like OpenAI, Anthropic, CrewAI, and LangChain, must integrate telemetry signals recording an agent’s persistent identity, any human interventions, and any changes to an agent’s permissions and memory. Meanwhile, enterprises adopting these agent frameworks have an influential role to play; enterprises in regulated industries, like banks and hospitals, can put pressure on the agent developers to supply these missing observability records where standards do not yet exist. Enterprises must also define policies, permissions, and roles that set out what agents can safely do, since telemetry alone cannot assess whether agents broke specific rules. Finally, regulators and AI standards institutes must specify the observability records required for high-stakes AI deployments while leaving the technical details to industry. We have already seen initial momentum here: NIST’s agent standards initiative in the US is working on agent identity, one of the most salient gaps we found in our research. 

No bank would let a thousand unbadged employees loose on customer accounts and hope to piece together what happened when something goes wrong. Yet that’s how we’re deploying AI agents. Making their behavior observable is the first step toward fixing this. And by building the regulatory and policy ecosystem that makes it possible to act on what we see, we can hold organizations accountable. Until then, every task we hand to an agent hands over our agency, too. Madhulika Srikumar is the Head of the AI Safety Program at the Partnership on AI, a global non-profit partnership of leading industry, academic, and civil society organizations, advancing best practices in AI governance. Claire Leibowicz is the Director of AI, Trust, and Society at the Partnership on AI and a doctoral candidate at Oxford University.

In association withAtlan Across a business, any number of AI platforms, systems, and tools may be at work simultaneously—for example, AI-powered customer platforms, conversational assistants, and workflow and process automation tools. Each one is sophisticated, advanced, and quick to respond. But, when set to work, the outcomes and actions that emerge may be wrong, confused, or lacking a proper understanding of the business. Many organizations find themselves in this situation—and it could be one of key reasons they struggle to extract returns on their AI investments. An MIT Technology Review Insights poll finds that while every organization surveyed is using AI today to some degree, fewer than one in five (16%) are extracting transformative value from their deployments. The largest percentage of respondents (40%) say they are seeing moderate gains in specific tasks or teams, while 21% report meaningful gains, leaving 23% that are seeing minimal returns or none at all. Companies are mostly still piloting AI use cases (according to 38% of respondents) or using it to assist employees’ work with tools like co-pilots or chat assistants (35%). Just 12% of enterprises have AI agents in production and 3% say they are using AI to do most of the work in key processes. “There’s a big gap between what the models are capable of and how the models are actually being useful inside organizations,” says Prukalpa Sankar, co-founder of Atlan. “For every real-life intent and purpose, artificial general intelligence is already here. It’s just not really useful in most companies today.”
The missing layer, Sankar says, is context, or how well AI understands a business. A context layer is the architectural infrastructure that sits between an organization’s business systems (for example, systems of record and data, and its AI platforms). It translates enterprise intelligence into governed, machine-usable context for AI agents or models and helps knowledge compound. “Without a context layer for AI, it’s almost like you’re hiring the smartest intern in the world, blindfolding them, and throwing them into your organization,” Sankar says. Download the report This content was produced by Insights, MIT Technology Review’s custom content arm, not its editorial staff. It was researched and written by humans, with any AI tools that may have been used limited to production processes under human oversight.

Today, we’re announcing the next frontier model: Gemini 4 Carbon, our strongest model with state-of-the-art cybersecurity and long horizon capabilities.Carbon ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59. It shows significant gains across coding and agentic tasks as well as across key capabilities critical for cybersecurity defense, landing a firm lead in SEC-bench Pro (88.2%) for finding vulnerabilities, ExploitGym (54.8%) for exploitation, and CWE-bench (71.0%) for patching. On internal benchmarks, Carbon drives a strong uplift in generating secure code, and in live coding environments with our own teams it shows reduced misalignment and fabrication rates.Building on Gemini 4 Argon, which was launched two weeks ago, Carbon illustrates the pace at which we’re pushing the frontier forward. Our researchers and models are working hand in hand to advance capabilities faster, while strengthening our safeguards in tandem.We’re taking a deliberate approach to rollout and are actively engaged in the U.S. government’s voluntary process for pre-release model testing. Starting this week, we will make Carbon available to a subset of cyber defenders and will gradually phase access to additional trusted partners through our Fairwind Program. We’re continuing to gather feedback and iterate on guardrails before making Carbon available to developers, businesses, and consumers via our product surfaces as soon as possible.Gemini 4 Carbon will launch at the same introductory price

as Argon at $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price.