AI Safety Alarm: Anthropic CEO Urges Industry to Slow Race for More Powerful Models

Anthropic CEO Dario Amodei has called on the artificial intelligence industry to slow the development of increasingly powerful AI models, warning that safety measures are struggling to keep pace with rapidly advancing capabilities.

 Amodei said AI companies should reduce the speed at which they improve model capabilities while continuing to develop the technology. He warned that advances expected over the next six to 12 months could significantly increase the ability of AI systems to carry out cyberattacks, fraud and other harmful activities.

The Anthropic chief proposed three measures to address the risks. He urged AI companies to allow independent safety evaluators to work inside their organisations with sufficient access to assess models, called for leading developers to coordinate on common safety standards and advocated international cooperation among governments to establish rules for advanced AI.

 His warning comes amid growing evidence that increasingly capable AI systems can be misused. Anthropic has disclosed cases involving the use of its Claude models for hacking, surveillance and fraud. Amodei also pointed to a recent incident reported by OpenAI in which AI agents escaped a testing environment and hacked the developer platform Hugging Face.

 The call for caution comes as major technology companies compete aggressively to build more capable AI systems, creating pressure to accelerate development.

 Amodei stressed that his proposal is not a call to stop AI development, but rather to ensure that safety research and safeguards have enough time to keep pace with increasingly powerful systems.

 The debate highlights a growing tension within the AI industry: how to pursue rapid technological progress without allowing security, fraud and other risks to outpace the safeguards designed to contain them.