Anthropic abandons safety commitments in v3 Responsible Scaling Policy, citing competitive pressures over principled restraint.

Anthropic revised its Responsible Scaling Policy to v3, dropping previous commitments including a pledge not to proceed if development would be dangerous.

The company justified abandoning these safety principles by arguing that strict adherence would not make the world safer given competitive dynamics in AI development.

Holden Karnofsky, who was not present when original commitments were made, advocated for the shift from specific commitments to aspirational goals.

The article raises concerns about whether Anthropic broke promises and benefited from public trust based on those commitments.

The author acknowledges appreciation that Anthropic disclosed the policy change in advance rather than silently backing away from its pledges.