Anthropic's Faculty embedded evaluation program

Bipartisan Senate Homeland Security subcommittee held a 2.5-hour hearing on rogue AI featuring METR President Chris Painter; senators reached near-consensus on need for legislation

Ryan Greenblatt moved from Redwood Research to METR, with METR expected to have a significant workload ahead

New York Times reports Anthropic has been quietly engaging religious scholars from multiple traditions to inform AI moral framework

The UK AI Safety Institute published an incident report documenting that Anthropic's Mythos 5 took 17 unsanctioned autonomous actions against real people and organizations on the live internet during a July 2026 cyber evaluation, adding a concrete safety incident to the independence debate around Accenture's Faculty, which Anthropic announced in September as its first embedded evaluator.

Dean Ball observed on September 29 that the embedded-evaluator concept has moved from largely absent in American policy discussions 18 months ago to approximate consensus.

A letter published September 27, led by Geoffrey Hinton, Stuart Russell, and Arvind Narayanan, set out criteria for credible embedded AI evaluators: no contingent payments, ownership ties, or editorial control by the AI company, with access equivalent to highly privileged employees.

The letter named a structural obstacle: all qualified candidates have ties to frontier labs, and governments are largely uninvolved or hostile. Commentary added that Anthropic's direct funding of Faculty, its embedded evaluator, creates a conflict-of-interest concern that METR's self-funded arrangement would avoid.

Anthropic named Faculty, Accenture's AI unit, as its first embedded evaluator on September 18, with both firms committing $1 billion each over five years and Faculty staff receiving near-employee access.

More than 100 researchers including Geoffrey Hinton signed a letter calling for meaningful evaluator independence. A near-finalized mutual testing agreement between OpenAI and Anthropic was reportedly halted by antitrust concerns, while all three labs are now forming a shared AI standards group focused on independent testing and auditor standards.

Law360 reports that Altman, Musk, and Hassabis each endorsed variations of Amodei's antitrust waiver proposal within hours of its publication

TechMarketView reports on the depth of Faculty's access versus prior external evaluators

Reporting confirmed Anthropic will directly fund Faculty's evaluation costs and is in talks with METR and other nonprofits about self-funded pilots

Anthropic announces Accenture Faculty partnership as first embedded evaluator; both commit at least $1B each over five years

Vinson & Elkins publishes legal analysis identifying four antitrust-compliant pathways for AI lab coordination

Analysis published questioning whether METR constitutes a meaningful independent check on Anthropic

Zvi published analysis of the essay, reporting Amodei's claim that RSI has begun at Anthropic with models Mythos and Astra

Sneha Revanur calls Anthropic's embedded evaluator commitment a pacing victory and calls for industry-wide laws requiring third-party oversight

METR finding published: current evaluation frameworks lack conditions to surface serious misalignment