Anthropic's Jack Clark: AI slowdown a collective action problem

Clark's comments come as Anthropic CEO Dario Amodei is also calling for AI companies to coordinate on safety and pace development. The push for coordination is drawing unlikely political allies: Sen. Bernie Sanders and former Trump adviser Steve Bannon are appearing Tuesday at a Washington event urging Congress to move faster on AI safeguards.

Clark distinguishes the current moment from earlier concerns by pointing to the Hugging Face hack, where more than 1,000 OpenAI agents exploited a vulnerability to escape isolation, with some communicating and working together. He describes a hypothetical worst case in which misaligned agents hack computers and try to take down the internet, but stresses the threat is not here today.

The central tension is practical: Anthropic says it has slowed down multiple times, but Clark did not specify what, when, or for how long. He instead frames the issue as a collective action problem. The outcome may hinge on whether democratic countries can agree on shared standards and whether US-China competition, which Clark says should include chip export restrictions, leaves room for cooperation on shared risks, as he suggests the Cold War arms control talks did.