Who audits the auditors? Mega-thread maps financial ties between AI safety evaluators and the six White House Accord labs

kevinnbass · x · 2026-10-17

kevinnbass published a mega-thread (with visualizations and a public repo) challenging the "independent external auditor" requirement in the September 29 White House Accord on Super Intelligence signed by Google, Anthropic, Meta, OpenAI, xAI and Nvidia, arguing the evaluator ecosystem is financially, ideologically and personally entangled with the labs it audits.

Key evidence from the thread:

27 documented Anthropic investors also hold stakes in at least 9 evaluator-adjacent companies (Amazon, Google, Nvidia, Sequoia among them); Macroscopic Ventures invested in Anthropic, Apollo Research's seed round and Gray Swan AI, and funds Redwood Research.
The $10M+ AI Safety Fund launched by Anthropic, Google, Microsoft and OpenAI granted to FAR.AI, Apollo Research, Nemesys Insights, SecureBio, OpenMined and Faculty; OpenAI's nonprofit gave CAIS $333,333 in each of 2023 and 2024.
Payment policies diverge: METR and Redwood take no lab money; Apollo charges market rates; CAIS caps for-profit AI developer revenue at 10%.
Epoch AI lists paid evaluations of Grok 4 and Gemini Deep Think commissioned by xAI and Google DeepMind; OpenAI commissioned and owns most of FrontierMath's 300 questions; Daniela Amodei was once listed as an Epoch leadership advisor.
Government testers are in the same network: Jason Matheny and Paul Christiano moved between Anthropic's trust, US CAISI and OpenAI's Foundation board; CAISI has documented ties to all six signatories.
The author proposes SEC-style auditor independence rules: name the auditor, disclose money, people and business ties. Of 8 adverse findings, 3 were followed by later funding and 5 were not — small counts, not dispositive.

Original post →

Who audits the auditors? Mega-thread maps financial ties between AI safety evaluators and the six White House Accord labs

More from Companies & People