METR’s Independence Questioned After X User Highlights Financial Links Between Company And Anthropic

An X thread from self-described independent researcher Kevin Bass has questioned METR's independence, alleging that Anthropic investor Dustin Moskovitz holds Anthropic stock through Good Ventures and that Good Ventures money flows via Open Philanthropy into AI safety nonprofits including METR. METR, formally Model Evaluation and Threat Research, says it enforces a "hard independence rule" barring funding from frontier AI companies and that its roughly $71 million in commitments raised over six months earlier this year came from philanthropic foundations and individuals, though it acknowledges receiving model access and tokens from OpenAI, Anthropic, and xAI. Bass's specific dollar figures and the exact funding lines to METR and the Tarbell Center come from his own self-published research and have not been independently verified, while former White House AI advisor David Sacks has publicly disputed METR's independence from Anthropic.

METR has become the AI industry’s default answer to a hard question of who could possibly keep tabs on AI labs. When Anthropic CEO Dario Amodei laid out his “pacing the frontier” proposal https://officechai.com/ai/anthropic-ceo-dario-amodei-says-ai-development-must-slow-down-to-pace-the-frontier/ for slowing down frontier AI development, the one concrete commitment in it was to give outside evaluators — Amodei specifically named METR — “employee-like access” inside the company, with desks, badges, and the ability to publish findings without Anthropic’s editorial control. That proposal has now pulled METR into a very public argument about whether it actually is the independent party it’s assumed to be. The thread The dispute was sharpened by a widely shared X thread https://x.com/kevinnbass/status/2099621874279817638?s=20 from Kevin Bass, who describes himself as an independent researcher, arguing that METR sits inside what he calls “the Anthropic Network” — a cluster of AI safety and AI-risk-focused organizations he says are financially entangled with Anthropic’s own success. Bass’s central claim is that Dustin Moskovitz, an early Anthropic investor, holds Anthropic stock through Good Ventures, the foundation he funds alongside his wife Cari Tuna, and that Good Ventures money flows — via Open Philanthropy and related vehicles — into the wider ecosystem of AI safety nonprofits, including organizations that fund AI-risk journalism such as the Tarbell Center. Because that stock has appreciated sharply as Anthropic’s valuation has climbed, Bass argues, the entire network has a financial stake in Anthropic’s success and can’t credibly serve as a neutral outside check on it. He posted a GitHub repository laying out the figures behind the claim and has said further evidence is coming. It’s worth being precise about what’s independently confirmed here and what isn’t. That Good Ventures and Open Philanthropy are major funders across the AI safety nonprofit world, and that Moskovitz has been a long-time Anthropic backer, is well established. The more specific claims in Bass’s thread — particular dollar figures on the stock stake, and the exact funding lines said to connect Good Ventures to METR and to Tarbell specifically — come from Bass’s own self-published research rather than from disclosed grant records or statements by the organizations named, and haven’t been independently verified. METR’s response and its funding rules METR has long argued that this kind of criticism misreads how it operates. The nonprofit, formally Model Evaluation and Threat Research, says it has a “hard independence rule”: it won’t accept funding from frontier AI companies and bars donations that are directed by staff of those companies. When it raised roughly $71 million in commitments over six months earlier this year, that money came from philanthropic foundations and individuals rather than AI companies, and METR pointed to that funding composition as evidence of its independence, arguing that an evaluator funded by the companies it evaluates would face an obvious conflict of interest. The organization does acknowledge working closely with the labs it evaluates — companies including OpenAI, Anthropic, and xAI provide it with model access and tokens for evaluations, research, and engineering — but says that is distinct from accepting money from them. That distinction hasn’t satisfied everyone. Critics, including former White House AI advisor David Sacks, have publicly disputed that METR is truly independent of Anthropic given the broader web of relationships between the company and the effective-altruism-adjacent safety community METR emerged from, and have argued that evaluator selection should come from organizations with no lab affiliation at all. Others have pushed back on the specific financial version of that argument: one review of the pacing plan’s aftermath concluded that no governance, funding, or conflict-of-interest records had actually been located to substantiate the claim that METR is compromised, as opposed to simply adjacent to a small, tightly networked field. Why it matters now The timing is what’s given the argument fresh weight. Amodei’s proposal envisions embedded evaluators playing something close to a regulatory role — reviewing training pipelines and processes, not just finished models, and helping decide when the industry should slow down. METR is also the organization Anthropic regularly cites in its own model announcements — its “time horizon” benchmark has become a standard reference point for how autonomously Claude models can work. If frontier labs and regulators lean on the same handful of evaluators to certify safety, the question of whether those evaluators are structurally independent, or simply well-regarded within a small and interconnected field, stops being an abstract one. Neither METR, Good Ventures, nor Anthropic has issued a direct, on-the-record response to Bass’s specific allegations as of publication. Bass has said additional evidence, including supplementary figures, is forthcoming.