The Download: Why AI's latest breakthroughs and fears may be more hype than reality

MIT Technology Review's daily newsletter The Download dedicated its September 22 edition to what it calls a summer of AI hype. The issue leads with an essay 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, who argue that the loudest claims of recent months resemble marketing more than genuine breakthroughs.

What the authors describe

The timeline starts in late April, when Anthropic said its model Claude Mythos finds software vulnerabilities better than most security experts. It continued with the OpenAI and Hugging Face hacking incident, after which Anthropic and Meta disclosed similar episodes of their own. Anthropic then claimed a mathematical breakthrough, and OpenAI quickly claimed one too. Most recently, Anthropic engineer Jacob Coxon drew wide attention by saying the company and OpenAI are heading toward self-improving superintelligence and putting human lives at risk.

What independent reviews found

Cybersecurity specialists say the hacking story is mostly about OpenAI's negligence and its failure to apply established security practices, not about models going rogue. Mathematicians who were initially stunned by OpenAI's claims about its chatbot Astra later concluded that the results were less novel than they first appeared, and accused the company of research misconduct and plagiarism. Tristan Buckmaster, a professor at New York University's Courant Institute, published a statement suggesting OpenAI had taken other researchers' work and misattributed it.

Why the framing matters

Gebru and Bender argue that calling these systems superintelligence or rogue models attributes agency to products rather than to the companies that build them. That framing sells the products as superhuman while helping their makers avoid accountability. The authors cite a statement signed by hundreds of mathematicians warning that the industry has a strong commercial incentive to overstate its capabilities, and asking policymakers to consult independent experts instead of press releases. They add that the industry dismisses public opposition to data centers as a distraction, while the real concerns are electricity bills, water use and air pollution.

Other stories in the same edition

The newsletter also reports that 22 countries have called for a new global body to oversee AI, with pre-deployment testing and common safety standards, while the US and China have not joined the declaration. Texas has halted new data center permits pending a grid audit, and California has passed laws covering power and water use. AMD became the twelfth company to reach a $1 trillion valuation as AI chip sales lifted its shares by more than 180% this year, and Meta's AI agent Muse overtook ChatGPT to top the US App Store.