The pause Anthropic deleted

Anthropic's Responsible Scaling Policy had one clause with teeth. Whenever a model's capabilities outran the company's ability to contain them, Anthropic promised to pause — categorically, no judgment call required. In version 3.0, effective February 24, that clause is gone.

If one developer paused to implement safety measures while others moved forward without strong mitigations, the policy now argues, that 'could result in a world that is less safe.'

What replaces it is a pause that fires only if Anthropic's own leadership decides two things are true at once: that the company is the front-runner in the AI race, and that the danger has become 'material.' Both are self-assessed. The hard commitments tied to capability thresholds become a 'Frontier Safety Roadmap' of targets the policy itself calls nonbinding. Independent reviewers describe the shift plainly: from commitments to statements of intent.

Anthropic states the rationale openly rather than burying it. One lab pausing while rivals race ahead, the argument goes, doesn't make the world safer — it just hands the lead to whoever cares least. So restraint is reframed as a coordination problem no single company can solve alone. The lab that built its identity on going slower has written down, in its own governance document, why going slower stopped making sense.

This matters beyond one company. The RSP, first published in 2023, was the template the rest of the industry copied — OpenAI's Preparedness Framework and Google DeepMind's Frontier Safety Framework both followed its shape. The strongest version of the genre just removed its load-bearing pre-commitment, and did so amid record fundraising rather than any crisis. The bio- and chemical-weapon safeguards stay; the oversight roles stay. What left is the promise that any of it could ever force the company to stop.