The head of Anthropic, an American artificial-intelligence company, has urged the AI industry to slow down so that safety can keep pace. On September 12, 2026, Dario Amodei warned that without a slowdown, AI could within six to 12 months lead a swarm of agents able to take over the entire internet.

His warning reflects growing concern that increasingly powerful systems might act on their own and cause serious harm. In a post on his website, he argued that even one or two extra years could improve safety and lower the risk of something going badly wrong.

Pressure has been building as employees resign and accuse their companies of racing ahead irresponsibly. A former safety researcher at Anthropic said many colleagues want to act responsibly but feel trapped in a race to build superintelligence. Another argued that the leading firms are "gambling with our lives."

Amodei proposed that all leading AI companies give outside evaluators ongoing access to monitor their safety work, which Anthropic plans to do itself. Sam Altman, the CEO of ChatGPT-maker OpenAI, quickly agreed to the same step. He has also said his firm will delay selling shares until after 2026. Elon Musk, a technology entrepreneur, added that Amodei was right.

Some critics, however, dismiss such warnings as a way to build excitement around the industry. Amodei still believes AI could bring huge benefits, such as cures for major diseases, but said its power to improve itself must be handled carefully. He concluded the industry has a duty to try, for the good of everyone.