Anthropic chief executive Dario Amodei has publicly called for the AI industry to slow the pace of model development, warning that the risks from increasingly capable systems are growing faster than the safeguards designed to contain them.
In a blog post published on Saturday, Amodei was careful to distinguish a slowdown from a halt. He is not calling for a freeze on model training or an end to technical progress.
What he wants is more time between capability milestones, enough space for companies to better align their models with human goals and values, and to subject them to rigorous independent evaluations before deployment.
Amodei estimated that pulling back before models reach critical capability thresholds could yield an additional one or two years to improve alignment work, which he said could reduce the risk of serious failures by a meaningful margin.
He acknowledged that not all of this can be done unilaterally. Some safeguards, he said, are within individual companies’ reach. Others would require coordination across the AI industry and, ultimately, between governments worldwide.
The statement is notable given Anthropic’s position as one of the leading frontier AI developers, competing directly with OpenAI, Google DeepMind and others in the race to build more powerful systems. A call for restraint from inside that race carries different weight than one from outside it.
