What changed
Nine governments and ten AI companies signed a declaration supporting common frontier-model evaluation and risk-management methods, standardized publication of non-sensitive findings, independent evaluation, fail-safe shutdowns, incident reporting, early-warning cooperation and protected whistleblower channels.
Why it matters
The declaration aligns state and developer expectations around comparable evaluations, operational shutdown authority and serious-incident reporting. It is a coordination signal, not an enforceable replacement for national law or independent audits.
What to do
Map current evaluation reports, shutdown paths, incident classes and protected reporting channels to the declaration. Record gaps, but wait for named standards, repositories, legal duties and implementation dates before treating commitments as controls.
Caveat
This is a non-binding diplomatic declaration signed on 9 November and published on 10 November, so the original event date is retained as a catch-up. It does not define thresholds, enforcement, timelines or a completed shared repository.