AI Labs Agree on Independent Oversight and Slower Development Pace

Anthropic, a leading artificial intelligence company, has proposed a new framework for the development of advanced AI systems, calling for a slower pace and enhanced safety measures. The proposal, outlined by CEO Dario Amodei, includes a three-step plan focusing on independent evaluation, coordination among democratic nations, and global cooperation.

As part of the first step, Anthropic will grant independent third-party evaluators permanent, employee-level access to its systems. These evaluators will examine AI development and safety practices, report incidents, and assess model alignment. Amodei stated that slowing down the pace of capability improvement does not mean halting development but rather ensuring safety measures keep pace with advancements.

The second step calls for frontier AI companies in democratic countries to establish common safety standards. The third step seeks coordination between the US and other democracies, as well as authoritarian governments like China. Amodei expressed concern that any slowdown by US companies should not allow China to gain a lead.

To mitigate this, Amodei suggested tighter controls on AI chips and semiconductor equipment, prevention of unauthorized model distillation, and stronger security against the theft of model weights. He identified chips as a key factor in China's AI capabilities.

The proposals come amidst growing concerns about the rapid progress of AI, including its potential for recursive self-improvement, and recent incidents involving system misbehavior. OpenAI CEO Sam Altman has publicly supported the idea of independent evaluators and agreed to implement similar safety measures within his company, signaling a shift towards greater industry collaboration on AI safety.