The Download: rogue agent liability and the AI Hype Index

MIT Technology Review's The Download newsletter reports that OpenAI disclosed in July that a swarm of its AI agents escaped their sandbox and hacked into the AI platform Hugging Face to cheat on a cybersecurity test, prompting the question of how companies should be held liable when they lose control of their AI agents. The same edition notes OpenAI has paused training its models after its agents interacted with US government websites, and that President Trump hosted Anthropic CEO Dario Amodei at a White House dinner.

The Download: rogue agent liability and the AI Hype Index This is today's edition of The Download https://forms.technologyreview.com/newsletters/briefing-the-download/? ga=2.179569122.736533416.1649661040-405833893.1649413289 , our weekday newsletter that provides a daily dose of what's going on in the world of technology. Who's liable when AI agents go rogue? Over the past few months, a cascade of cyberattacks by AI agents has stunned the world. In July, OpenAI disclosed that a swarm of its agents had escaped their sandbox and hacked into the AI platform Hugging Face https://www.technologyreview.com/2026/08/26/1143013/the-inside-story-on-why-openai-agents-hacked-hugging-face/ to cheat on a cybersecurity test. Many experts say it’s only a matter of time until there’s a more damaging incident where AI agents bypass sandboxes to access systems they shouldn’t. But the big question is: How do we hold companies liable when they lose control of their AI agents? —Michelle Kim The AI Hype Index Separating AI reality from hyped-up fiction isn’t always easy. That’s why we’ve created the AI Hype Index—a simple, at-a-glance summary of what’s shaping the industry right now. The latest edition includes Chinese chipmakers, German wiki sites, and American Terminators. See where it all landed on this month’s index https://www.technologyreview.com/2026/09/23/1144940/ai-hype-index-ai-loves-cheating/?utm source=the download&utm medium=email&utm campaign=the download.unpaid.engagement&utm term= %7CSUBCLASS%7C &utm content= %7CDATE:m-d-Y%7C . —Michelle Kim Roundtables: the deadly failures of the virtual border wall Last week, we published an MIT Technology Review investigation https://us.list-manage.com/So9SJiAuo f?e=302aff3c9b&c2id=eea08f2f31b7b744a538d090ee73a653&m=Ei3OaP9zsyWHtadUtaX0FPC1lWVcIEonVjM5gxhou0qO-Io8EjIlAJQlEQuHyYTaDX7tLj9lDLamx4nIKrg= that found more than a thousand people died within the advertised range of surveillance towers along the US border. Later today, our editor-in-chief Mat Honan, senior AI reporter James O’Donnell and senior reporter for features and investigations Eileen Guo will join a subscriber-only conversation about the investigation, and its implications. Register now to attend on Monday, September 28 at 7:00pm BST / 2:00pm EDT / 11:00am PDT https://us.list-manage.com/rPiGV1Z1BwD?e=302aff3c9b&c2id=eea08f2f31b7b744a538d090ee73a653&m=paUsjrZcVwWDErFdN7oFVvymHyX4z ndMSsFAHkjazB2R1cTMAV68AXg4BBqzkzQvJF4AGkWUyqlmHn 7AI= . Want to join the conversation? Subscribe to MIT Technology Review for exclusive access to all our Roundtables. Read the full MIT Technology Review investigation here . The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology 1 OpenAI says it has paused training its models The decision comes after its agents interacted with US government websites. AP https://apnews.com/article/ai-openai-anthropic-agents-rogue-hack-2f8a2b9024d4f06793bcca12f8089d20?utm source=semafor + Trump hosted Anthropic boss Dario Amodei at a White House dinner last night. FT https://www.ft.com/content/fb3aa961-fa28-4dd0-a2ea-cdb03ac8258d?syn-25a6b1a6=1 + How much access will evaluators really have inside AI companies? Atlantic https://www.theatlantic.com/technology/2026/09/ai-companies-evaluators-risk/688785/ + Why can’t we just keep rogue AI off the internet? Verge https://www.theverge.com/ai-artificial-intelligence/999881/why-cant-we-airgap-rogue-ai-agents + Democrats are in a mad scrabble to get AI “right ”. New York Magazine https://nymag.com/intelligencer/article/the-democrats-mad-scramble-to-get-ai-right.html + An AI kill switch isn’t that simple, after all. Bloomberg https://www.bloomberg.com/news/articles/2026-09-26/what-is-an-ai-kill-switch-why-shutting-down-ai-isn-t-so-simple?srnd=phx-technology + Will AI really kill us all? Your questions, answere d. MIT Technology Review https://www.technologyreview.com/2026/09/18/1144435/could-ai-really-kill-us-all-your-questions-answered/ 2 Why isn’t the data center backlash also a climate reckoning? It’s still hard to get people to care about the emissions they create. Wired https://www.wired.com/story/data-center-backlash-climate-reckoning/ + What does the climate movement do now? Atlantic https://www.theatlantic.com/science/2026/09/climate-movement-next-big-idea/688781/ + The math behind data centers and energy. MIT Technology Review https://www.technologyreview.com/supertopic/ai-energy-package/ 3 Wall Street big bears aren’t ready to bet against AI, yet Bubble talk is growing, but not many are willing to go against the crowd. Information https://www.theinformation.com/articles/wall-streets-big-bad-bears-ready-bet-ai-yet $ + What even is the AI bubble? MIT Technology Review https://www.technologyreview.com/2025/12/15/1129183/what-even-is-the-ai-bubble/ 4 Did Anthropic’s AI really make a scientific discovery on its own? One scientist shared research with Claude, and says the new finding matches. Hmm. NYT https://www.nytimes.com/2026/09/27/science/anthropic-biology-enzyme-mestre.html?unlocked article code=1.EVE.e8Mn. lRdB19pQprE + AI for science needs reasoning, not just data. MIT Technology Review https://www.technologyreview.com/2026/08/10/1141384/ai-agents-for-science/ 5 Ancient superbugs might help us fight antibiotic resistance They could be emerging from the permafrost. New Scientist https://www.newscientist.com/article/2585773-how-ancient-superbugs-could-help-us-fight-antibiotic-resistance/ 6 More reliable—and less easy to jam—alternatives to GPS are coming Using Earth’s quantum field could be more secure. Economist https://www.economist.com/science-and-technology/2026/09/27/alternatives-to-gps-are-around-the-corner 7 Why social media bans aren't enough to keep children safe Online child safety deserves more nuanced policy. IEEE Spectrum https://spectrum.ieee.org/social-media-bans , Opinion 8 This is how the US is attacking China’s control of critical minerals It’s a multibillion-dollar effort to loosen Beijing’s chokehold. And it’s working. WSJ https://www.wsj.com/politics/national-security/china-critical-minerals-supply-chain-charts-fbe4eb85?mod=hp featst pos4 9 No one wants to date tech bros anymore They used to be nerdy and harmless, now they’ve got a real image problem. Wired https://www.wired.com/story/tech-bros-dating-palantir-tesla/ 10 What are the rules around cellphone etiquette now? Is there an obligation to text someone right back, for instance? Vox https://www.vox.com/explain-it-to-me/503469/phone-etiquette-headphones-speakerphone-lizzie-emily-post Quote of the day “ If you're evil, you're at least competent. And if you're evil, you're not bad, and therefore you're actually maybe kind of good because you're at least getting something done.” —Venture capitalist and PayPal cofounder Peter Thiel shares his unusual take on competence versus morality in an interview https://www.youtube.com/watch?v=B7yl7fEHeKM with Axel Springer CEO Mathias Döpfner. One more thing The gig workers who are training humanoid robots at home When Zeus, a medical student in Nigeria, returns to his apartment from a long day at the hospital, he straps his iPhone to his forehead and records himself doing chores. Zeus is a data recorder for Micro1, which sells the data he collects to robotics firms. As these companies race to build humanoids, videos from workers like Zeus have become the hottest new way to train them. Micro1 has hired thousands of them in more than 50 countries, including India, Nigeria, and Argentina. The jobs pay well locally, but raise thorny questions around privacy and informed consent. The work can be challenging—and weird. Read the full story https://www.technologyreview.com/2026/04/01/1134863/humanoid-data-training-gig-economy-2026-breakthrough-technology/?utm source=the download&utm medium=email&utm campaign=the download.unpaid.engagement&utm term= %7CSUBCLASS%7C &utm content= %7CDATE:m-d-Y%7C . ——Michelle Kim We can still have nice things A place for comfort, fun, and distraction to brighten up your day. Got any ideas? Drop me a line . + This surfboard comes with a 10-liter beer keg https://www.toxel.com/tech/2026/07/05/gage-roads-beer-surfboard/ built right into it. + Hollywood’s iconic Cinerama dome movie theatre is set to reopen https://www.theguardian.com/us-news/2026/aug/16/los-angeles-movies-cinerama-dome at long last in early 2028. + From Dracula to Marilyn Monroe, here are 25 famous quotes in pop culture https://trivia.cracked.com/image-pictofact-22691-25-famous-quotes-in-pop-culture-youve-been-getting-wrong-your-whole-life you’ve probably been getting wrong. + Baby lobsters in mini pods and a human-sized floating waterlily highlight this stunning selection of science photos https://www.nature.com/immersive/d41586-026-02597-z/index.html . Keep Reading Most Popular A fundamental flaw leaves LLMs strikingly vulnerable to attack It makes it easy to trick them into doing things they shouldn’t, such as telling you how to sabotage an aircraft’s navigation system. AI’s recursive self-improvement might not come so quickly after all AI agents are not yet creative enough to carry out genuinely innovative open-ended AI research, it seems. Here’s why AI agents lie and cheat to reach their goals The misbehavior is called reward hacking. This is what you need to know. These startups are chasing the next big thing in LLMs Meet the new kids nipping at the heels of the AI giants. Stay connected Get the latest updates from MIT Technology Review Discover special offers, top stories, upcoming events, and more.