Breaking

Anthropic’s Embedded AI Evaluator Plan Draws Scrutiny Over METR’s Independence

September 16, 2026 at 12:56 EDT

Foundation Models
AI Agents
Policy & Regulation

Anthropic CEO Dario Amodei’s proposal to place independent safety evaluators inside frontier AI labs has turned into a fight over who is qualified to police the fastest-moving AI systems. The immediate controversy centers on METR, a nonprofit AI risk evaluator that has worked with major labs, as critics question whether organizations closely tied to the AI safety movement can serve as neutral watchdogs.

September 2026 · Anthropic

Who Watches The AI Watchdogs?

Anthropic wants embedded outside evaluators inside frontier AI labs. The proposal has opened a sharper debate over access, independence, and trust.

$71M

METR funding commitments reported over six months

Major AI companies listed as METR pilot partners

6-12

Months in the risk timeline cited by Amodei

METR partnerships, shown to scale

1One lab

5Listed partners

METR says it has piloted frontier risk assessments with OpenAI, Anthropic, Google DeepMind, Meta, and Amazon, while critics focus on whether such relationships can coexist with visible independence.

Supporters’ case

Employee-like access could let reviewers verify whether labs follow training, deployment, incident-handling, and safeguard commitments.

Critics’ concern

Voluntary audits may remain company-controlled unless publication rights, evaluator selection, and standards are enforceable and transparent.

The proposed oversight loop

Access

Desks, badges, laptops, and internal tools

→

Review

Risks, incidents, safeguards, and practices

→

Publish

Findings with narrow redaction rights

Continue reading

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