Why this summer's AI hype falls apart under scrutiny

In an opinion piece published on September 22 by MIT Technology Review, Timnit Gebru, executive director of the Distributed AI Research Institute (DAIR), and linguist Emily M. Bender argue that this summer's wave of AI announcements does not survive expert scrutiny.

A summer of loud claims

The authors trace the sequence: at the end of April, Anthropic claimed its model Claude Mythos could find software vulnerabilities better than most security experts. Then came the OpenAI–Hugging Face hacking incident, after which Anthropic disclosed a similar case "proudly" and Meta "reluctantly". Next, Anthropic said one of its models had made a mathematical breakthrough, and OpenAI soon claimed one of its own. Most recently, Anthropic engineer Jacob Coxon went viral announcing his departure, saying the two companies are "racing straight towards self-improving superintelligence and gambling with our lives".

What the experts found

The fanfare drew intense coverage, the authors write, while the picture that emerges once experts examine the events gets far less attention. Cybersecurity specialists describe the hacking cases mainly as OpenAI's negligence and failure to adopt basic, established security practices — not "models gone rogue". Mathematicians first stunned by a press release claiming the chatbot Astra had solved problems open for at least a decade later concluded the results were not as novel as first appeared and accused the company of research misconduct and plagiarism. Two days before OpenAI claimed another mathematical breakthrough, NYU Courant Institute professor Tristan Buckmaster published a statement suggesting the company had taken other people's work and attributed it improperly.

"Superintelligence" as marketing and as a shield

Claims of imminent, dangerous superintelligence are not grounded in good science or engineering, the authors argue, but in transhumanism, eugenics and wishful thinking about imagined digital humans. Calling systems "superintelligence" or "rogue models" gives agency to products rather than to the companies building them — marketing them as superhuman while helping the firms dodge accountability. A statement signed by hundreds of mathematicians warns of "a strong commercial incentive... to overstate the capabilities of their products" and urges policymakers to consult experts rather than press releases. The piece also criticizes the industry's framing of anti-data-center activism as a "distraction": the public is asked to fear a fictional machine god more than data centers' climate impact, asthma among nearby residents, rising electricity bills or water diverted to cooling. The authors conclude that the best outcome of this summer of hype would be for policymakers and the public to keep their skepticism and recognize the pattern next time.