Gebru and Bender say AI's summer hype outran the evidence
MIT Technology Review's daily newsletter The Download leads with an opinion piece by Timnit Gebru, executive director of the Distributed AI Research Institute (DAIR), and Emily M. Bender, a professor of linguistics at the University of Washington. They write that it has been a busy few months for AI hype, with companies making breathless claims about hacking incidents, mathematical breakthroughs and the prospect of self-improving superintelligence. In each case, the authors say, massive fanfare from the companies involved, presented as mea culpas in the hacking cases, was met with intense press coverage; but once experts had time to examine what actually happened, very different stories emerged. Gebru and Bender argue there is a strong commercial incentive for companies to overstate AI capabilities, and that the illusion of speed and urgency the industry promotes is also a ploy meant to misdirect policymakers and the public. Their conclusion: readers should remain skeptical of AI hype.
The same edition's must-read roundup covers several other AI stories. Twenty-two nations have called for a new global body to oversee AI, wanting pre-deployment testing and common safety standards, though neither the United States nor China is part of the new declaration; separately, OpenAI has proposed its own global AI safety standards and wants the US to lead the international effort. Texas and California have both moved to rein in data centers, with Texas halting new permits pending a grid audit and California passing new laws covering data centers' power and water use. Chipmaker AMD has become the twelfth company to reach a $1 trillion market capitalization, with AI chip sales helping its shares rise more than 180% this year. Meta's AI agent Muse has overtaken ChatGPT to top the US App Store and helped give Meta shares their best month in 13 years, even though Muse carries a major zero-day vulnerability and Amazon has blocked it from shopping on its site.
The digest's quote of the day comes from Ben Casselman, The New York Times's chief economics correspondent, who explains why people feel trapped in a lose-lose dynamic with the AI boom: if AI succeeds it may kill jobs, and if it fails the economy could suffer along with people's retirement savings and jobs anyway. The rest of the edition rounds up shorter items: a new edible battery designed to power medical devices inside the body, tested in pigs with an RFID tracker and stomach stimulator; a move that could give Trump's budget chief veto power over NIH grants; an epigenetic-editing treatment for hepatitis B that has reached human trials; UK biotech funding for tree vaccines and gene-edited butterflies aimed at helping nature adapt to climate change; a rocket shortage making it harder to launch satellites as SpaceX winds down Falcon 9; and a mysterious "component issue" affecting Dyson's $500 toothbrush. A separate feature describes a new wind-powered cargo sailboat, inspired by traditional vessels used in the Marshall Islands, that made its maiden voyage last year and could cut cargo shipping emissions by up to 80% compared with a fuel-powered ship; cargo shipping is responsible for about 3% of the world's annual greenhouse-gas emissions.
Key facts
- Timnit Gebru (DAIR) and Emily M. Bender (University of Washington) argue AI companies have a strong commercial incentive to overstate their capabilities, using the illusion of speed and urgency to misdirect policymakers and the public.
- 22 nations have called for a new global body to oversee AI, with pre-deployment testing and common safety standards; neither the US nor China is part of the declaration, and OpenAI has separately proposed its own global AI safety standards.
- Texas has halted new data-center permits pending a grid audit, while California has passed new laws covering data centers' power and water use.
- Chipmaker AMD became the twelfth company to reach a $1 trillion market cap, with AI chip sales helping its shares rise more than 180% this year.
- Meta's AI agent Muse overtook ChatGPT atop the US App Store and helped give Meta shares their best month in 13 years, even though it carries a major zero-day vulnerability and Amazon has blocked it from shopping on its site.
Why it matters
The piece is a direct pushback on the AI industry's own narrative: two researchers with long records of criticizing overhyped AI claims argue that this summer's headline-grabbing stories about hacking, math breakthroughs and self-improving superintelligence look much smaller once independent experts check them. Their broader point is not about any one claim but about incentives: companies benefit commercially from urgency and spectacle, and that same urgency shapes how regulators and the public respond. The same day's news bears out how high the stakes of that narrative are, from a 22-nation push for global AI oversight to AMD's trillion-dollar valuation built substantially on AI chip demand.
Who it affects
Policymakers weighing the 22-nation proposal for pre-deployment testing and safety standards, and the US and China, whose absence from that declaration is noted without explanation. AI companies whose hacking, mathematical-breakthrough and superintelligence claims are the direct target of the critique, though the article does not name which companies or claims it means. AMD and its investors, Meta and users of its Muse agent, and residents of Texas and California affected by new data-center rules.
How to use it
Treat sweeping AI announcements, especially ones framed as breakthroughs or as companies' own mea culpas over hacking incidents, as claims to be checked once independent experts have had time to examine them, rather than as settled fact on release day. The piece itself does not name the specific incidents it is referring to, so readers looking to verify a particular claim need to trace it back to the original coverage rather than to this newsletter.
How solid is it
This is a signed opinion piece, not a neutral report: Timnit Gebru is DAIR's executive director and Emily M. Bender is a University of Washington linguistics professor, and both have made careers of scrutinizing AI-industry claims, so the piece should be read as their argued position. It does not name the specific hacking incidents, mathematical breakthroughs or superintelligence claims it says turned out to be overstated, so its central claim cannot be checked against specifics from this text alone. The rest of the digest, including the AMD market-cap figure and the Meta App Store ranking, is reported as straightforward news drawn from other outlets.
Risks and caveats
The article does not name the specific companies or claims behind the hacking, mathematical-breakthrough or self-improving-superintelligence cases it calls hype. It gives no detail on what the 22 nations' proposed pre-deployment testing and common safety standards would actually require. It says neither the US nor China is part of the new declaration but does not explain why, or whether either plans to join later. And it gives no timeframe or location for the Meta Muse zero-day vulnerability it mentions.
“If AI succeeds, then it may kill all of our jobs. If it fails, then the whole economy falls apart, and your 401(k) blows up, and maybe you still lose your job.”
— Ben Casselman, The New York Times's chief economics correspondent