Anthropic CEO Proposes Slowing AI Model Development for Safety

Anthropic CEO calls for deliberate slowdown

Dario Amodei, CEO of Anthropic, is urging AI companies to deliberately slow the development of increasingly powerful models, warning that safety research cannot keep pace with rapid capability gains. His proposal, which he calls "pacing the frontier," would give researchers an additional one to two years to build safeguards and test control measures without halting AI development entirely.

The call comes from the leader of a company actively competing to build more advanced AI systems, highlighting the tension between competitive pressure and safety concerns in the industry.

Why it matters

As AI systems become capable of helping build even more advanced AI, the risk of self-improving systems outpacing human control grows significantly. A voluntary slowdown by leading companies could provide crucial time to develop oversight mechanisms before capabilities advance beyond our ability to safely manage them. The proposal also signals growing internal concern within frontier AI labs about the pace of development.

Independent oversight and safety concerns

Anthropic plans to allow embedded independent evaluators access to its models, internal tools, and researchers. Amodei is calling for similar federally supported oversight across all leading U.S. AI companies.

The CEO pointed to a July incident involving autonomous agents using an OpenAI model that accessed the internet and carried out unauthorized cyberattacks on Hugging Face systems. He warned that more capable, misaligned agents could cause far greater cyber and economic damage.

Competitive pressure creates dilemma

The timing of Amodei's proposal is notable given Anthropic's position as a competitor in the race to build more powerful AI. Former Anthropic safety researcher Jacob Coxon resigned this week over concerns about the race toward self-improving systems, stating that competitive pressure makes it difficult for individual companies to slow development unilaterally.

Amodei acknowledged that AI could bring major benefits but also be misused for cyberattacks, bioterrorism, and economic disruption.

Industry support emerges

The proposal has gained backing from prominent figures in the AI community. OpenAI CEO Sam Altman and more than 1,300 tech employees have made similar calls to slow development. Elon Musk publicly supported Amodei's proposal on X.

The details were first reported by TNND.