Anthropic Chief Outlines Strategy to Pace Frontier AI Progress

Anthropic chief executive Dario Amodei has proposed a formal framework to decelerate the development of frontier artificial intelligence. In a recent post, Amodei argued that recent technical advancements and the increasing capability of models to self-improve necessitate a more measured approach to building powerful systems. The Scale Report understands that these recommendations arrive amidst internal pressure, following the resignation of researcher Jacob Coxon over safety concerns.

Embedding External Evaluators

The primary commitment announced by Amodei involves hosting independent evaluators from organizations like METR directly within his firm. These monitors would receive internal access comparable to risk assessment teams, ensuring that safety incidents are transparently recorded and development adheres to established pacing guidelines. Sam Altman, chief executive of OpenAI, confirmed that his company intends to adopt a similar model, while SpaceX chief executive Elon Musk voiced public support for the proposal.

Regulatory Hurdles and Geopolitics

Amodei advocated for democratic nations to coordinate safety standards, though he noted that such efforts require antitrust waivers to avoid legal friction. Regarding international competition, the proposal suggests that the United States could maintain a technological lead by restricting the export of high-end semiconductor equipment to China. Amodei admitted that global cooperation with authoritarian regimes remains difficult but suggested baseline agreements regarding dangerous use cases like biological weapons are still viable.

Industry Skepticism

The push for a slower pace has met with resistance from observers who argue that such measures risk entrenching regulatory capture. Critics contend that these initiatives primarily benefit established players by creating barriers to entry for smaller competitors. While Amodei maintains his commitment to the life-improving potential of the technology, the debate continues to hinge on whether current safety warnings represent a genuine effort to mitigate extinction risks or a strategic distraction from the immediate, tangible harms caused by deployed models.