Anthropic, OpenAI Dial Back Safety Language as AI Race Accelerates
Anthropic has revised its Responsible Scaling Policy, dropping a key safety pledge that previously required fully implemented safeguards before training advanced AI models. This move marks a shift in the company’s approach, now allowing continued development even if some risk mitigations aren’t fully in place. Leadership cites the rapid AI advancement by competitors as a main reason for abandoning unilateral safety commitments. Anthropic’s policy change comes amid a dispute with the U.S. Defense Department over restricted access to its Claude AI system, separating it from other industry players like Google and OpenAI. This reflects a broader trend, as OpenAI also recently removed explicit references to safety in its mission statement. Experts note that early “AI safety” terminology envisioned risks differently from those posed by current large language models, and industry language is evolving to emphasize transparency and economic competition over definitive safety promises. Anthropic’s updated policy now highlights transparency measures and risk reporting, pledging to delay development only for significant catastrophic risks. The shift occurs in the context of intense fundraising, competition, and political pressure within the AI sector.
