The fight is over who gets to check whether AI companies’ systems behave as intended. One side wants established consulting and accounting firms. The other wants nonprofit and civil-society evaluators, many from the AI safety community. According to a Washington Post report by Nitasha Tiku, Ian Duncan and Gerrit De Vynck, published October 6, 2026, the Trump administration and its allies have adopted some safety proposals while trying to sideline the advocates who made them.
What the dispute is about
The Post reports that AI safety advocates’ warnings helped draw attention to the technology’s risks. The administration and its allies have since embraced some of those proposals. They have also sought to push the advocates themselves out of the process.
As an Amazon Associate I earn from qualifying purchases.
A recent accord signed by President Trump and AI chief executives sharpened the question of who should audit company systems. The Post’s account of the accord is the only description available here. Its full text, binding force and audit requirements are not detailed, so this article makes no claims about them.
Recommended Free Tools
One concrete result: Anthropic announced a partnership with Accenture to build a team of outside evaluators. That came after David Sacks criticized the independence of METR, a nonprofit evaluator that works with Anthropic.
#1 Best Overall
Who is saying what
The critics and skeptics
- David Sacks, an investor and former senior White House AI adviser, said: “In my view they should be professional auditors, not these NGOs.” He questioned whether METR is independent given its work with Anthropic.
- Emil Michael, the Pentagon’s chief technology officer, reportedly criticized calls for action against extreme AI scenarios as promoting “extinction, death-cult-like philosophies.”
- Nathan Leamer, director of Build American AI, said he had briefed congressional staff and conservative figures about effective altruism. He called it antithetical to his Christian values. The Post says Build American AI is linked to Leading the Future, a political campaign group funded by Andreessen Horowitz and OpenAI President Greg Brockman.
- An anonymous administration official told the Post: “There has absolutely been a rhetorical push within certain corners of the White House and broader Trump administration to get the message out about effective altruism and its potential dangers.”
The advocates and evaluators
- Brad Carson, co-founder of Americans for Responsible Innovation, argued that Trump allies are trying to discredit people who want meaningful AI regulation. He also said: “In D.C., at least, and maybe in the broader culture, we won that fight. The vibe is with us.”
- Rayan Krishnan, CEO of Vals AI, described his Washington outreach this way: “It almost felt like a new version of McCarthyism.”
- Miranda Bogen of the Center for Democracy & Technology, who directs its AI Governance Lab, said: “Where we find ourselves now is trying to determine who will actually be in the room in making decisions.”
- Adam Gleave, CEO of FAR.AI, argued that excluding the safety community would shut out relevant expertise: “When it comes to looking at millions of agent transcripts and trying to piece together what they did and how they were interacting, I don’t think there are many organizations in the world that have that expertise.”
In between
Judd Rosenblatt, CEO of AE Studio, said people in the administration recognize a need for third-party assessments but are wary of effective altruism. That suggests the objection is partly to the movement’s identity and not to outside testing as such. The Post’s reporting is consistent with that reading, but it is Rosenblatt’s characterization, not an established finding.
The corporate-auditor side
Chamath Palihapitiya said on the All-In podcast that 8090 was building audit infrastructure with EY. EY said it was working with 8090 on capabilities to improve visibility, traceability and human accountability in consequential AI-enabled decisions and workflows. That is EY’s own description, and it is narrower than “auditing AI companies’ models.”
Rank #2
The reputational attack on the safety movement
The Post describes criticism of effective altruism and the AI safety community as cultish or unserious. Whether this amounts to a coordinated campaign is contested. Carson and others frame it that way. The Post attributes that characterization to them and does not independently establish it. The anonymous official’s remark confirms a rhetorical push in parts of the administration, but it does not show coordination beyond that.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsNonprofit evaluators versus consulting firms
The Post does not run a side-by-side test of the two models, so no verdict on which is better is established. The questions below are useful ways to judge any auditor. They are not findings from the report.
Rank #3
| Question | Why it matters | What the report shows |
|---|---|---|
| Independence | Evaluators paid by or partnered with the firm they assess may soften findings | Sacks questioned METR’s independence over its Anthropic work. Anthropic’s Accenture partnership raises the same question for a commercial firm. The report does not resolve either. |
| Relevant technical experience | Frontier models and agents are hard to evaluate | Gleave says few organizations can review millions of agent transcripts. This is his claim, not a measured comparison. |
| Transparency | Outsiders need to see methods and results | Not stated in the report |
| Accountability | Auditors need professional or legal consequences for failures | Sacks’s preference for professional auditors rests on this kind of argument, though the report does not spell it out |
| Conflict handling | Both models have funding and client ties | Not stated in the report |
What the report does and does not establish
- The dispute is real and reported on the record by named people. It concerns who evaluates AI systems and whether safety nonprofits get a role.
- Claims that nonprofit auditors lack independence, and claims of a coordinated smear campaign, are positions held by one side or the other. Neither is settled.
- Some advocates acknowledge the need to prove their legitimacy, according to the report, even as they argue that shutting out existing expertise would weaken oversight.
- The report offers no independent statistics on auditor performance. A proposal for “two-month boot camps” came up as a conference idea, not as evidence of what training works.
The practical stake is Bogen’s point about who is in the room. If audits go mainly to consulting firms, safety nonprofits lose influence over how systems are tested. If they stay central, the administration’s allies will keep pressing questions about their independence. What would settle it is published methods, disclosed funding and conflict rules, and evidence of how well each type of auditor performs. None of that is in the report.
Quick Recap
Rank #4
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
