Will METR announce a third-party evaluation protocol for Anthropic models by 12 November 2026?
Posted 2026-09-13 — 60 days before the deadline, stated before the outcome.
METR is an independent research organization that develops methodologies for assessing the capabilities and risks of advanced AI systems. Anthropic is a technology company that develops large language models and related AI products. Formal evaluation protocols are mechanisms designed to provide standardized, external verification of model performance and safety.
The claim
METR will announce or confirm the activation of a formal third-party evaluation protocol for Anthropic models by 2026-11-12.
Reasoning trace
- Dario Amodei explicitly pledged METR access to Anthropic models for safety adherence verification in a public statement on 2026-09-12.
- METR has recently established credibility as an independent investigator of rogue agent incidents, making it the logical candidate for this role.
- Implementing a third-party evaluation requires defining technical criteria and access mechanisms, which typically takes several weeks to finalize.
Evidence so far
10 confirming · 0 denying signals · 4 verified · 1 broken assumptions. The evidence itself is part of the full dossier.
Probability history
Updated 19 times since mint (last on 2026-10-22) — 22% at mint → 76% today.
The full dossier behind this prediction — the evidence trail, the risk analysis, the monetization scenarios — is available on request: [email protected].
Probabilities are calibrated against the system's own resolved history; after the deadline the outcome is resolved and audited — including the calls we get wrong. Nothing on this page is investment advice.