Anthropic CEO Calls for Slower AI Development as Safety Risks Grow
Anthropic CEO Dario Amodei has urged leading artificial intelligence companies to slow the development of increasingly powerful models, arguing that safety testing and external oversight are not keeping pace with rapidly advancing capabilities. His proposal calls for independent evaluators, shared industry standards and international cooperation as concerns grow over the misuse of AI in cyber operations, surveillance, fraud and weapons-related work.
Amodei outlined a three part framework designed to give developers more time to manage the risks created by increasingly capable models. His proposal calls for independent evaluators to receive deep access to frontier AI companies, common safety standards across the industry and international cooperation on advanced AI risks.
The appeal followed Anthropic's latest threat intelligence report, which detailed misuse cases involving cyber operations, surveillance, fraud, biological activity, conventional weapons development and attempts to copy model capabilities. Anthropic said the cases covered activity it disrupted between December 2025 and August 2026.
The company stressed that the examples represented notable or novel threats rather than typical use of Claude. Even so, the report shows how capable systems can help malicious actors move faster or operate at greater scale.
Amodei is not asking companies to stop training models. He argues that developers should moderate the speed at which capabilities are advanced while giving safeguards, alignment work and external testing time to catch up.
OpenAI CEO Sam Altman and xAI chief Elon Musk publicly supported parts of the proposal. Altman said independent evaluators with access similar to employees were a strong idea and that OpenAI would adopt the approach.
Coordinating a slowdown will be difficult in a market shaped by large infrastructure investments, intense competition and possible public listings. Amodei also said safety cooperation in democracies would need to account for competition with China, including controls around advanced chips, model weights and distillation.
The proposal shifts the industry's safety debate from whether risks exist to whether rival companies can coordinate before the next generation of models arrives.
Source: Reuters, 12 September 2026
