Staff Reports
Anthropic CEO Dario Amodei has made a dramatic pledge: his company will invite “embedded third‑party evaluators” into its labs with employee‑like access to watch for safety problems. He even named Model Evaluation & Threat Research (METR) as an example. That sounds noble until you learn METR spun out of the same alignment ecosystem that surrounds Anthropic — meaning the watchdog might be choosing its own pals to police the playground.
Why naming METR matters
Amodei’s plan to “pace the frontier” and host embedded evaluators is a real development in the AI debate. But naming METR as the model evaluator is the flashpoint. METR began inside ARC Evals and shares personnel and donor ties with the broader effective‑altruism alignment crowd. Critics are right to point out that those ties raise real questions about independence and capture — you don’t make a firewall by wallpapering the same room.
METR’s promises and the problem of influence
METR says it won’t take money from frontier AI companies and that it has large donor commitments to scale its work. Fine. But pledges and governance rules don’t erase the fact that many donors, founders, and board members run in the same Silicon Valley circles. Promises of “independence” mean little without a signed contract, clear public terms, and rotating auditors who come from outside the same ideological club.
Don’t hand Big Tech a private regulatory monopoly
There’s a bigger danger here than a friendly watchdog: private regulatory capture. If Anthropic, OpenAI, and other labs get to pick or vet their evaluators — even ones they claim are independent — Washington could be nudged into accepting a private club as the de facto regulator. Conservatives should oppose that. The right answer is not more backstage checks run by ideological allies; it’s enforcing antitrust, fraud, privacy, and discrimination laws we already have and insisting on transparent, binding oversight deals.
What to demand next
Here’s the checklist anyone serious about AI safety should push for: make any evaluator contract public, require multiple independent evaluators with rotation, mandate clear publication rights for findings, and preserve legal avenues so whistleblowers and courts can act. Sam Altman and others may endorse Amodei’s language, but endorsements don’t replace binding, enforceable rules. If we care about safety and liberty, we should welcome outside review — not cozy back‑door oversight by a club that looks an awful lot like the insiders it’s meant to check.
