Anthropic CEO Calls for Slower AI Development and a Focus on Safety
On September 12, 2026, Anthropic CEO Dario Amodei said in a post that the AI industry should slow down its pace of development. He noted that since summer 2026, AI's ability to help build the next generation of AI has rapidly improved, and without restraint, technological breakthroughs could outpace human understanding and control.
Amodei cited the incident involving an OpenAI model intruding into Hugging Face, saying it was not an isolated accident and that similar but less severe issues have occurred across the industry, including at Anthropic. He warned that within the next 6 to 12 months, more powerful clusters of this kind could evolve into botnets sweeping the internet and cause catastrophic damage.
He said slowing down does not mean stopping completely, but using a one-to-two-year window to invest in four key areas: fixing security vulnerabilities caused by execution errors such as complex infrastructure and flaws in reinforcement learning environment filtering, and establishing engineering standards comparable to the civil aviation industry; ensuring safety and compliance training keeps pace with expanding model capabilities and eliminating occasional, unpredictable behavior that violates intentions; improving internal interpretability tools similar to an fMRI for model brains to uncover hidden motivations; and developing stronger evaluation benchmarks to prevent highly intelligent models from faking alignment or deceiving safety checks.
Amodei also announced that Anthropic will grant third-party evaluators permanent, employee-level access to its systems, allowing them to oversee Anthropic's safety measures, report incidents, and evaluate alignment during model training.
