Breaking

Anthropic's METR Safety-Evaluator Plan Draws Backlash Over Independence and Politics

September 16, 2026 at 12:14 EDT

Policy & Regulation
AI Agents
Security

Anthropic CEO Dario Amodei's call to embed outside AI safety evaluators inside frontier labs has drawn a political backlash, with critics questioning whether METR is too culturally and ideologically aligned with the AI safety movement to serve as an independent check. The dispute turns a technical governance proposal into a broader fight over who gets to verify the safety of the most capable AI systems.

September 2026 · Anthropic

Embedded AI evaluators face a credibility test

Anthropic wants outside teams such as METR inside frontier labs to verify safety practices. The debate now centers on whether those evaluators can be seen as independent enough.

steps in Amodei's pacing framework

weeks in Anthropic's METR research agreement

$71M

METR commitments raised in six months

Anthropic hacking cases disclosed this year

The proposal is a three-part sequence

3 steps

Embedded evaluators, democratic coordination, global coordination

Supporters' case

Embedded reviewers could make safety commitments verifiable and give outsiders visibility into incidents, training pipelines and internal controls.

Critics' concern

Skeptics question whether evaluators linked to the AI safety community can be viewed as neutral watchdogs for frontier labs.

The unresolved issue

The fight is no longer only about model behavior. It is about who gets access, who writes the standards, and whether voluntary oversight can earn public trust.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.