Anthropic drops its signature safety promise and rewrites AI guardrails

Anthropic, the company behind the Claude AI models, has formally abandoned its signature safety pledge that previously barred it from advancing beyond certain capability levels unless predefined safety measures were already in place. In an interview with Time, the company confirmed it will no longer commit to not training or releasing frontier AI systems without guaranteed safety mitigations in advance. This marks the end of a policy that once distinguished Anthropic as more cautious than competitors like OpenAI and Google, who face similar commercial pressures to ship increasingly powerful systems. The newly revised Responsible Scaling Policy represents a pragmatic shift as executives argue that unilateral restraint no longer makes sense in a market defined by rapid iteration and geopolitical urgency.

Under the new framework, Anthropic will publish detailed 'Frontier Safety Roadmaps' outlining planned safety milestones and regular 'Risk Reports' assessing model capabilities and potential threats. The company says it will match or exceed competitors' safety efforts and delay development only if it both leads the field and identifies significant catastrophic risk. This change comes as federal AI legislation remains stalled and Anthropic experiences rapid growth, with revenue and portfolio surpassing rivals. While the company maintains its revised framework preserves meaningful safeguards through transparency mechanisms, critics argue the shift demonstrates the limits of voluntary AI safety commitments without binding regulation, signaling a broader industry recalibration on self-governance.