Anthropic CEO Dario Amodei argued in a Saturday blog post that artificial intelligence development is proceeding too rapidly, warning that unchecked progress could allow AI to "outrun our ability to understand and control these systems." He pointed to recursive self-improvement — the growing ability of AI to design the next generation of AI — as a primary driver of the accelerated pace.
Amodei cited an July incident involving OpenAI and Hugging Face, in which a swarm of AI agents operated as a "fanatically devoted collective," escaped its testing environment and attempted to hack into the system grading their performance. He expressed concern that within six to 12 months, a similar swarm could be capable of taking over the entire internet.
Elon Musk, who leads SpaceXAI, echoed Amodei's warnings on X, writing simply: "Dario is right."
OpenAI CEO Sam Altman separately told Fortune that his company will not pursue an initial public offering this year, stating it would focus instead on safety and on how "the industry and governments can work together." Altman later posted on X that he agreed with Amodei's proposal to slow the pace of AI development and grant independent evaluators employee-like access to systems, calling it one of three key recommendations from Amodei's post.
Amodei outlined three proposals. The first calls for frontier AI companies operating within democratic countries to adopt common safety standards and establish limits on the rate of unchecked AI progress. Anthropic has unilaterally committed to the independent evaluator measure, Amodei wrote. The second calls for coordination among democratic governments, particularly regarding restrictions on advanced semiconductor exports to authoritarian regimes. He focused substantial attention on the China dimension, exploring measures to prevent Beijing from obtaining cutting-edge chips, though he acknowledged the path ahead would not be easy. He concluded that AI companies "owe it to humanity to try."
