Anthropic asks frontier labs to prepare for a coordinated pause

Anthropic has called for the world's leading AI developers to create a coordinated and verifiable way to pause advanced-model development if risks grow too quickly. The company says autonomous task capability has been doubling roughly every four months and warns that future systems could help improve their own successors faster than institutions can respond. The proposal is not an immediate unilateral halt, and rivals including OpenAI, Google, Meta, xAI, and Mistral have not committed to it. Critics may view the move as safety positioning from a company racing toward an IPO, but the proposal still raises a concrete governance question: whether competing labs can agree in advance on evidence thresholds, verification methods, and enforcement strong enough to stop a race during a genuine emergency.

Key details: June 5, 2026, Anthropic, Coordinated and verifiable pause proposal, Autonomous task capability doubling roughly every four months, No rival commitments announced.

Continue swiping for more AI Brief stories.