Don’t Be Fooled by This Summer of AI Hype

Summary

In this opinion essay, Timnit Gebru and Emily M. Bender argue that a series of highly publicized AI events was presented as evidence of imminent artificial general intelligence, even though later expert scrutiny produced a less dramatic account. Anthropic said its Claude Mythos model outperformed most security experts at finding software vulnerabilities, while OpenAI, Anthropic, and Meta disclosed hacking-related incidents involving their models. The authors say cybersecurity experts viewed the incidents primarily as failures to apply established security practices, particularly in OpenAI’s case, rather than as evidence of rogue models or autonomous systems building civilizations. They also discuss OpenAI claims that its chatbot Astra achieved major mathematical breakthroughs. Mathematicians later said the first result was less novel than initially presented and accused the company of research misconduct and plagiarism; a separate statement by NYU professor Tristan Buckmaster alleged that OpenAI had improperly taken credit for others’ work. The authors argue that claims of rapidly approaching self-improving superintelligence are ideological narratives, not conclusions supported by sound scientific or engineering practice. They say programming and mathematics are especially useful for marketing such claims because outputs can be evaluated more readily than many real-world tasks. A statement signed by hundreds of mathematicians calls on policymakers to consult independent experts rather than rely on corporate press releases. The essay further argues that terms such as “rogue models” shift agency away from companies and away from issues including negligence, unauthorized data use, plagiarism, and the environmental and public-health costs of data centers. It concludes that policymakers and the public should slow down, seek independent expertise, and treat corporate urgency as a possible form of marketing rather than proof of technological transformation.