Three weeks after Anthropic chief executive Dario Amodei called for a slower, independently checked AI race, Shattered.io says the public record still shows no named evaluator or published process from the companies that endorsed the idea. The gap is between agreeing with a safety pledge and showing how it works.
Watch Desk analysis
What happened
Amodei’s 12 September essay proposed three steps: give independent evaluators employee-like access to frontier AI labs; have companies coordinate on shared safety standards; and pursue international government cooperation. Shattered.io’s 2 October review says it found no public confirmation from Anthropic or OpenAI naming an evaluator, start date or operating protocol for the first step.
The review also describes a public split over coordination. It says Meta chief executive Mark Zuckerberg rejected industry-wide coordination in a 26 September interview. The review does not establish what companies may be doing privately, and the absence of a public announcement is not proof that no work is under way.
Why it matters
The evaluator pledge was presented as something a company could begin on its own. That makes it the clearest part of the proposal to assess from the outside: a name, a start date and a description of access would turn a broad commitment into something readers could scrutinise.
The other two steps depend on competitors and governments agreeing to work together. Shattered.io’s account suggests those ambitions face a more visible obstacle: at least one major lab has publicly rejected industry-wide coordination.
Our read
The useful question is no longer simply whether AI leaders will endorse safety in principle. It is what evidence they are willing to put in public. A named evaluator and a clear account of its access would be a better measure than another round of agreeable statements. Shattered.io’s review is a snapshot of the public record, not a window into private arrangements.
What to watch
- Whether Anthropic or OpenAI names an evaluator and explains its access.
- Whether either lab publishes a start date, reporting cadence or operating protocol.
- Whether other major AI companies join, or reject, the call for industry coordination.
Discussion spark: Should frontier AI labs have to publish the name, access and findings of independent evaluators to make safety pledges credible, or would that reveal too much about their systems?
Sources and evidence
- AI Pacing Pledge: 0 Evaluators Named in 20 Days [2026] – shattered.io (2 October 2026, 05:47 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.