Anthropic has abandoned its core safety pledge that once defined its Responsible Scaling Policy (RSP), a dramatic shift that experts warn signals industry-wide retreat from safety commitments.

The company announced it will no longer require itself to guarantee adequate safety measures before training new AI models, a reversal of its 2023 promise that positioned it as the most safety-conscious AI lab. This decision came after executives concluded unilateral safety pauses would be ineffective as competitors raced ahead, potentially creating a dangerous "race to the bottom" where weaker safeguards prevail.

Anthropic's new approach emphasizes transparency through Frontier Safety Roadmaps and quarterly Risk Reports, while committing to match competitors' safety efforts. Critics like METR's Chris Painter see this as triage mode for an unprepared society, noting the company's retreat from its founding premise that frontier AI development requires frontier safety research.

The policy overhaul coincides with Anthropic's massive $30 billion funding round valuing it at $380 billion, reflecting investor confidence in its commercial model over safety-first approaches. This move leaves the company less constrained by its own policies, raising questions about whether safety commitments can survive competitive pressures in the accelerating AI race.