Forget superintelligence: error-prone AI nearly sparked world war three this month

Guardian opinion piece (1 Oct 2026). Standfirst: "The risks of AI aren't what we think they are, as a recent security incident between China and the United States reveals." Gebru and Bender contrast the attention given to Anthropic engineer Jacob Coxon's resignation warning with the little attention paid to CNN's 18 Sep report. They list the policy reactions to Coxon's warning: Newsom's "AI kill-switch" mandate, Hochul exploring kill-switches, and the Sanders–Casar bill to "ban artificial superintelligence". The CNN report described a chatbot-generated intelligence report that nearly led US forces to board a Chinese ship. The authors note that the CNN report "hasn't been verified by other major outlets" and that CNN could not learn which chatbot product was used.

They argue that LLMs are "stochastic parrots" that are neither mysterious nor powerful for such uses. They write that "Claims that system behavior is mysterious arise from deliberate mystification on the part of those selling those systems", and that harms arise "because people believe that the products are superintelligent, not because they actually are". They call for regulating the use of "error-prone products that shouldn't be used in high-stakes scenarios", with evaluation within each use case and accountability kept with people. They cite Lina Khan's point that there is "no exemption from the law" for AI.

Bylines: Gebru is executive director of DAIR (her book Deep Unlearning: The Radicalization of a Tech Idealist is "set to publish on 16 February"); Bender is professor of linguistics at the University of Washington and co-author of The AI Con. Hacker News submission: 6 points.

Short quotes only (see Summary). Full text at the URL.