Anthropic Sets New AI Safety Baseline with External Model Audits

Anthropic has initiated external safety evaluations with METR, becoming the first frontier AI firm to grant an independent body access to its most advanced, non-public models. This move establishes a new baseline for responsible AI development, strategically positioning transparency as a competitive asset. In a landscape where rivals like OpenAI and Google face mounting scrutiny over their internal safety processes and high-profile team departures, Anthropic’s action is a direct challenge, shifting the debate from self-attestation to verifiable, third-party audits and setting a precedent that will be difficult for competitors to ignore. This partnership fundamentally alters the AI safety assurance model, creating a clear winner in the court of public and regulatory opinion: independently audited labs. For closed-source developers like Google and OpenAI, the pressure to follow suit is now immense, exposing the potential vulnerability of their less transparent safety claims. The losers are the proprietary, internal-only safety teams whose findings will now appear less credible. By providing METR with access to models that are "more capable than any model currently publicly available," Anthropic is weaponizing transparency, creating a defensible moat built on trust rather than just performance metrics. The critical variable going forward is the substance and severity of METR