Viral report alleges Anthropic’s AI safety watchdog conflicted

New research has accused Anthropic of using “AI doom” media narratives to pump Dustin Moskovitz’s equity in the company to fund foundations that support its safety evaluator and curiously positive safety reviews.

Substack author Kevin Bass made the claim in a lengthy X post on Monday. In the post, he also calls for a Congressional investigation into the alleged financial conflicts of interest at Anthropic’s AI safety evaluator. It has so far drawn nearly 5 million views on social media.

Protos has not evaluated its veracity.

Focusing on the equity relationships between Anthropic and Model Evaluation and Threat Research (METR), the entity that checks its frontier AI models for safety, the analysis prompted several observers to report “SBF flashbacks.”

According to Bass, Moskovitz’s Anthropic equity helps to fund Good Ventures Foundation and Coefficient Giving, which in turn support METR as Anthropic’s conflicted safety evaluator.

It also funds the Tarbell Center for AI Journalism, which pushes AI doom media to keep the flywheel spinning.

Anthropic CEO Dario Amodei is another curly-haired billionaire and Effective Altruist who has, like FTX’s Sam Bankman-Fried, constructed a web of self-serving entities.

These include:

• A household name with a high corporate valuation (Anthropic)
• Affiliated investment vehicles that almost no one has heard of
• Active political lobbying efforts
• Effective altruism-aligned nonprofits
• A well-funded media campaign to keep everything capitalized

Although no one is accusing Amodei of secretly stealing billions of dollars of customer deposits like Bankman-Fried, there seems to be a commingling of financial incentives between Anthropic and its ostensibly independent safety checker.

Anthropic installs ‘embedded evaluators’ for safety

Over the weekend, Anthropic CEO Dario Amodei proposed “embedded evaluators who have employee-like access to verify safety practices and report incidents.”

Boasting about his company’s supposed rigor, Amodei urged other frontier companies to “follow suit.”

Bass, in complete contrast to this narrative, recast that story as a payroll scandal.

Trace the capital back, he claims, and METR ultimately relies on Facebook co-founder Dustin Moskovitz, a self-professed effective altruist.

Forbes reported that he and his wife, Cari Tuna, moved an Anthropic stake into an unnamed nonprofit vehicle in early 2025.

By November 2025, Forbes estimated its worth at a staggering $500 million.

Forbes later bounded the donated holding up to 0.8% of Anthropic, a company that has raised money at a $965 billion valuation this May. At that valuation, 0.8% could be worth over $7 billion.

Moskovitz said, “Our Anthropic shares are entirely in our foundation — no personal benefit.”

Coefficient Giving CEO Alexander Berger wrote that the shares didn’t literally go to Coefficient Giving. However, a commenter disagreed with the spirit of that claim, noting that Amodei’s sister, Daniela, is married to Coefficient Giving co-founder Holden Karnofsky.

“If any of Coefficient Giving’s board members are large METR donors, that is a direct and obvious conflict of interest,” the commenter concluded.

The latest available tax filing for the other nonprofit connected to METR, Good Ventures Foundation, reports $10.1 billion in assets at that nonprofit.

The filing names hundreds of public stocks but puts private equity and venture capital into generic buckets. Anthropic doesn’t appear by name.

METR annualized funding commitments of $142M

From whichever entities ultimately originated most of their fortunes, METR announced $71 million of new commitments over the past six months.

Despite this large budget, METR safety evaluators claim to refuse Anthropic or any AI lab funding, although frontier labs provide substantial free tokens for testing.

Coefficient Giving is legally separate from Good Ventures, although Good Ventures serves as its founding partner. Coefficient Giving’s live archive records $3.7 million across two direct cash awards to Tarbell.

Tarbell acknowledges that Coefficient supplied most of its funding as of 2025, while maintaining that it doesn’t allow donors to control its reporting nor media efforts.

Anthropic announced an eight-week agreement with METR this month. The evaluator will get access to employees and internal transcripts.

METR’s conflict policy says it’s never received payment for company-identifying assessments, and it “strives to be supported by broad and independent funders.”