tech · 2023 to 2026

The Pause Nobody Takes First

A frontier lab offered to slow down, but only if rivals verifiably slowed too. That single condition is the whole game: a coordinated pause everyone might prefer, and no one will take alone.

Facts verified as of June 2026.

The sequencing is the story. In 2023 Anthropic's original Responsible Scaling Policy committed the company not to train or deploy a system unless it could guarantee adequate safety mitigations in advance, a unilateral pause pledge. On February 24, 2026 Anthropic released Responsible Scaling Policy version 3.0, removing that unilateral halt and reframing around collective action.

The stated rationale was a collective-action problem: because catastrophic risk depends on the actions of all frontier developers, unilateral slowing would leave Anthropic behind without reducing aggregate risk. The prior robust mitigations, the company said, might prove impossible to implement without collective action.

Play the same structure: The Pause Nobody Takes First.

Sources

[VERIFIED] Anthropic offered a conditional, verification-gated pause and stated more than 80 percent of merged code was Claude-authored. Anthropic, When AI builds itself (May 2026)
[VERIFIED] RSP v3.0 removed Anthropic's unilateral halt pledge and reframed around collective action. Anthropic, Responsible Scaling Policy Version 3.0 (February 24, 2026)
[VERIFIED] A formal pause agreement is a horizontal output restriction at high risk of per se illegality; unilateral pausing carries minimal risk. Lawfare, Can Frontier AI Labs Lawfully Agree to Pause (June 29, 2026)
[VERIFIED] OpenAI's reported position is that democratic governments, not private firms acting alone, should set AI rules. Al Jazeera, Anthropic urges AI labs to pause (June 5, 2026)