DeepMind backs Anthropic's push to pace AI

A rare moment of alignment at the frontier

When two of the most aggressive players in frontier AI start agreeing on safety, it is worth paying attention.

Anthropic CEO Dario Amodei just published an essay titled "We Must Pace the Frontier," arguing that the AI industry needs to deliberately slow down its release cadence. Amodei laid out a three-part plan, committing Anthropic to the first step immediately: giving third-party evaluators permanent, employee-level access to their systems so external experts can independently verify safety measures, report on incidents, and evaluate model alignment directly during training.

Google DeepMind CEO Demis Hassabis quickly backed the stance. Hassabis stated that Dario's essay points towards the right path forward, noting that while the operational details still need working through, the direction is correct for this critical moment. He also pointed back to DeepMind's own recent push for an industry-wide standards body focused on frontier AI.

Why it matters

Frontier labs are realizing that "trust us" is no longer an acceptable safety policy.

Anthropic's unilateral move to grant external auditors internal, employee-grade system access sets a high benchmark for operational transparency. DeepMind's endorsement makes sense: by tying Amodei's call to an industry-wide standards body, Hassabis is trying to turn individual voluntary commitments into enforceable, collective guardrails.

We think this alignment creates real pressure on rivals to match this level of external oversight. That said, getting the entire ecosystem to willingly tap the brakes—especially when commercial stakes are this high—remains the hardest puzzle yet.