Anthropic ditches its core safety promise
The 'safety-first' AI company replaces its hard commitments with public goals, citing competition and a $200M contract ultimatum.
Anthropic, the AI safety-focused company founded by ex-OpenAI researchers, has made a significant strategic pivot by abandoning its core, binding safety commitment. The company announced it is replacing its two-year-old Responsible Scaling Policy (RSP)—which mandated pausing development if model capabilities outstripped safety controls—with a new, non-binding 'Frontier Safety Roadmap.' This framework outlines public goals rather than hard commitments, marking a departure from Anthropic's founding principle of enforced self-regulation. The change is driven by the company's assessment that its previous policy failed to create an industry-wide 'race to the top' on safety and could hinder its competitiveness, especially as other AI firms advance rapidly without similar constraints.
The policy shift occurs amidst intense pressure from the U.S. Department of Defense, which gave CEO Dario Amodei a deadline to roll back AI safeguards or risk losing a $200 million contract and potential blacklisting. While Anthropic states the policy update is separate from Pentagon negotiations, the timing highlights the commercial and governmental pressures facing 'safety-first' AI developers. The new policy decouples Anthropic's internal safety plans from its industry recommendations, acknowledging that Washington's current anti-regulatory climate and the pace of AI advancement require a more adaptable approach. The company will now publish regular, detailed reports on its risk mitigation plans and model capabilities, framing this as increased transparency, even as it adopts a more flexible stance on development guardrails.
- Anthropic replaces its binding Responsible Scaling Policy with a non-binding 'Frontier Safety Roadmap,' removing the mandatory pause clause for unsafe capabilities.
- Faces a Pentagon ultimatum: roll back AI safeguards or lose a $200M contract and risk government blacklisting by Friday.
- Company argues that unilateral safety pauses could backfire, creating a 'less safe' world if less careful competitors advance unchecked.
Why It Matters
Signals a major retreat from enforceable AI safety principles under commercial and government pressure, setting a new precedent for the industry.