Anthropic's Accenture-Run 'Independent' Evaluation Sparks Conflict-of-Interest Backlash

Episode 1 · Anthropic-Funded Accenture "Independent Evaluation" Sparks Conflict-of-Interest Backlash (2026-09-19, 10 posts)

Anthropic announced an "independent evaluation" of its frontier AI models by Accenture, funded directly by Anthropic, drawing broad criticism from the AI safety community. Commenters argued Accenture's deep business ties to Anthropic compromise its independence, while safety researchers said the nonprofit evaluator METR would be far better suited to serve as an embedded auditor. Gary Marcus separately raised conflict-of-interest questions about METR staff holding Anthropic stock, which a self-identified METR insider rebutted, saying only about 2 METR employees previously worked at Anthropic.

Confirmed

• Anthropic announced the evaluation partnership with Accenture; firstadopter noted it is directly funded by Anthropic, and Accenture's large enterprise consulting business creates a conflict of interest for an "independent" assessment paid for by the party being evaluated
• PaulYacoubian cited Grok-compiled research showing deep ties: in December 2025 the two expanded a multi-year partnership establishing an Accenture Anthropic Business Group inside Accenture
• Hesamation criticized Anthropic's embodied-AI evaluation, where the "independent" evaluator is the same Accenture that co-built that joint business group and holds product cooperation and investment ties with Anthropic
• Safety researchers including nablatheta argued METR, a nonprofit focused on frontier-model evaluation, is far better qualified and more independent, urging Anthropic to commit to METR embedded audits rather than quietly dropping METR while publicly claiming third-party audit via Accenture
• Gary Marcus tweeted: "How many METR employees hold Anthropic stock? A fair question."
• CFGeek, identifying as METR-affiliated, responded that a rough estimate shows only about 2 METR employees previously worked at Anthropic, and noted the headcount overlap statistics Marcus amplified have methodological issues—former Anthropic employees who forfeited or sold shares would inflate the figure, while other cases could understate it
• A forwarded discussion offered a comparative view: Accenture is culturally more independent from Anthropic but less independent in incentives, while METR's transparency mission makes it more credible even when findings are unfavorable

Why it matters

• If AI companies' safety and capability evaluations are performed by commercially tied partners chosen and paid by the evaluated party, "independence" becomes meaningless—trusted third-party evaluation is a cornerstone of AI governance
• The dispute is ultimately an industry-wide institutional choice: should external audits go to specialized nonprofit evaluators or consulting firms
• The Marcus-CFGeek exchange shows that even challenges to evaluator independence need fact-checking—headcount and shareholding methodology directly shape conclusions

• METR, not Accenture, should be Anthropic's embedded auditor, argues AI safety observer — nabla_theta · 2026-09-19
• METR is far more qualified than Accenture to be Anthropic's embedded auditor — nabla_theta · 2026-09-19
• Accenture-Anthropic deep ties undermine claims of independence, post argues — PaulYacoubian · 2026-09-19
• Researchers Push Anthropic to Commit to METR Embedded Audits Over Accenture — nabla_theta · 2026-09-19
• Anthropic funds Accenture 'independent' frontier AI evaluation, criticized for conflict of interest — firstadopter · 2026-09-19
• Anthropic slammed for using business partner Accenture as "independent" evaluator — Hesamation · 2026-09-19
• Gary Marcus questions how many METR employees hold Anthropic stock — GaryMarcus · 2026-09-20
• METR rebuttal to Gary Marcus: only 2 staff ever worked at Anthropic — CFGeek · 2026-09-20
• Commenter flags overcount/undercount flaws in METR-Anthropic stockholder overlap estimate — CFGeek · 2026-09-20
• Debate over Anthropic safety evals: Accenture vs METR independence questioned — GaryMarcus · 2026-09-20

Episode 2 · Gary Marcus Slams Dario Amodei Over Conflicted AI Auditors (2026-09-19, 5 posts)

On September 19–20, AI commentator Gary Marcus published a series of posts accusing Anthropic CEO Dario Amodei of multiple credibility failures within a single week, centered on conflicts of interest among nominally 'independent' AI safety evaluators. The criticism remains Marcus's perspective, with no response from Anthropic in the available material, but it exposes a structural problem in AI safety evaluation worth attention.

Confirmed

• Marcus listed several inconsistencies: after Dario's call to slow down frontier AI, Anthropic pledged employee-level access for third-party evaluators, yet designated METR and Accenture as external overseers rather than genuinely independent bodies. When METR's closeness to Anthropic was questioned, Dario responded by striking an evaluation deal with Accenture, which Marcus argues merely swaps one conflict of interest for another, with no actual slowdown at Anthropic.
• Citing Anka Reuel's disclosure, Marcus highlighted that a self-described independent evaluation firm simultaneously announced a business deal with the entity it evaluates—where the deal's success depends on findings not being alarming, lest clients be scared off. Marcus called this 'utterly corrupt.'

Unconfirmed

• The specific nature and extent of the ties between METR, Accenture and Anthropic rest solely on Marcus's account and those he cites; no official statements from the institutions or Anthropic are present.

Why it matters

• The credibility of frontier AI safety evaluation depends on evaluator independence. Economic or social ties between evaluators and evaluated entities undermine neutrality, potentially reducing 'responsible scaling' pledges to formality. Marcus's critique brings this structural issue into public view and may shape how future third-party evaluation mechanisms are designed.

• Gary Marcus slams 'independent' AI evaluator over conflict of interest with the company it rates — GaryMarcus · 2026-09-19
• Gary Marcus: Dario picked auditors with ties to Anthropic and hasn't slowed down at all — GaryMarcus · 2026-09-19
• Gary Marcus: three ways Dario Amodei blew his credibility in seven days — GaryMarcus · 2026-09-19
• Gary Marcus lists three ways Dario Amodei blew his credibility in one week — GaryMarcus · 2026-09-19
• Gary Marcus slams Anthropic's Accenture eval deal amid METR independence concerns — mjdramstead · 2026-09-20