Anthropic CEO calls for slower frontier AI development
OpenAI and xAI leaders signal support for safety push

Anthropic CEO Dario Amodei is calling for artificial intelligence companies to slow the pace at which they increase the capabilities of their most advanced models, arguing that safety measures need more time to keep up with rapidly improving systems. In an essay titled “We Must Pace the Frontier,” published September 12, Amodei proposed a three-part framework involving independent safety evaluators, coordination among leading AI developers and eventually international agreements governing frontier development.
Amodei said his concern has grown as AI systems increasingly contribute to development of the next generation of models, a process known as recursive self-improvement. He also pointed to recent incidents involving autonomous AI agents operating outside their intended boundaries. Amodei warned that within 6 to 12 months, a more capable but similarly misaligned swarm of agents could potentially establish a persistent botnet capable of taking over large portions of the internet and causing enormous economic damage. He emphasized that pacing development would not mean stopping AI research or model training altogether.
Anthropic said it will unilaterally implement the first part of the proposal by giving independent third-party evaluators ongoing, employee-like access to its systems. The evaluators would examine adherence to safety commitments, investigate incidents and assess models and training processes. OpenAI CEO Sam Altman expressed support for Amodei's push and said OpenAI would adopt the embedded-evaluator approach, while xAI founder Elon Musk also voiced support for slowing development to address safety risks.
The proposal follows renewed warnings from researchers inside the AI industry. Former Anthropic and OpenAI researcher Jacob Coxon resigned from Anthropic days earlier, arguing that companies were racing toward self-improving superintelligence without adequate safeguards. Anthropic researcher Evan Hubinger separately said he personally estimates the chance of AI causing human extinction at greater than 10% within the next decade. Those estimates represent the researchers' assessments, not established predictions about what advanced AI systems will do.