Sources say Anthropic and OpenAI sought strict control of evaluators' reports after backing their independence
Anthropic and OpenAI sought significant control over what outside AI evaluators can publish, Reuters reported, citing two people familiar. Both companies dispute parts of that account. The fight turns on who decides which findings are trade secrets.
The Investor · Invest desk
Amodei backed publishing findings without editorial control. Anthropic reserves approval of new firms' reports. OpenAI seeks no approval of full reports. Contracts can bar confidential information without consent. Embedded evaluators may be kept from other labs. Faculty's terms are unsettled.
- What it sets up precedent Anthropic CEO Dario Amodei Said evaluators should have the right to publish key findings "without editorial control", claim 1
- Who has to decide decision Anthropic Reserves the right to comment on and approve reports from new evaluator firms, claim 11
- Who has to decide decision OpenAI Not seeking approval rights over full report text; evaluators can disclose major redactions, claim 12
- What it limits constraint Evaluators under lab contracts Contract language can bar publishing confidential information without the companies' consent, claim 3
- What it limits constraint Embedded evaluators Measures under discussion include barring them from working at another lab simultaneously, claim 6
- Who's exposed exposure Faculty Anthropic's first embedded evaluator; access and reporting details still being worked out, claim 7
| Who | How | Kind | Claim |
|---|---|---|---|
| Anthropic CEO Dario Amodei | Said evaluators should have the right to publish key findings "without editorial control" | precedent | 1 |
| Anthropic | Reserves the right to comment on and approve reports from new evaluator firms | decision | 11 |
| OpenAI | Not seeking approval rights over full report text; evaluators can disclose major redactions | decision | 12 |
| Evaluators under lab contracts | Contract language can bar publishing confidential information without the companies' consent | constraint | 3 |
| Embedded evaluators | Measures under discussion include barring them from working at another lab simultaneously | constraint | 6 |
| Faculty | Anthropic's first embedded evaluator; access and reporting details still being worked out | exposure | 7 |
What happened
- Last month, Anthropic CEO Dario Amodei argued that outside evaluators should be entitled to publish their main findings "without editorial control."
- On Oct 5, according to the two people, both labs were still seeking strict editorial control over what evaluators could publish.
- Anthropic told Reuters that the description did not fit all its discussions, and that evaluators it previously worked with control their methods, criteria, conclusions and what to publish.
- OpenAI told Reuters it was not seeking approval rights over the full text of reports and that evaluators can disclose whether major redactions were made.
Why it matters
- What doesn't add up contradiction Anthropic reserves a right to comment on and approve reports from new evaluator firms while they learn its models. That right sits against Amodei's "without editorial control." It also applies when a new reviewer has least leverage.
- Who's exposed exposure Evaluators bound by the contract language Reuters saw need the lab's consent before publishing anything the lab calls confidential, so the lab's definition of the word sets the limit.
- Who has to decide decision Faculty, Anthropic's first embedded evaluator, and any firm after it must accept terms while access and reporting rules are still being worked out.
- What it limits constraint Under the discussed bar on working at two labs at once, an embedded evaluator can serve 1 of the 2 labs at a time, or 50% of the pair.
Next story · Build · yesterday
Anthropic plans to put outside evaluators inside its labs, starting with AccentureConfirmed4 publishers
On paper, the public line and the private asks can both stand. OpenAI backed Amodei's position days later, saying assessors should have "editorial independence" [2], and a lab can honor that wording while still requiring consent before anything it calls confidential reaches the page. The labs have a stated reason to want the consent: according to Reuters' sources, they worry that evaluator staff working inside their walls could take proprietary information to rivals [4]. Having lab employees observe embedded teams as they work is one measure under discussion [5].
The fight is over where "confidential" ends. A separate person with knowledge of past dealings with the labs said that what they consider crucial intellectual property can overlap with information relevant to judging a model's risk. That overlap, the person said, leads to arguments and redactions [16]. Bars on sharing a client's confidential information are typical in technology contracts [9], so the clause itself is ordinary. Reuters reported that evaluator independence is a concern because Anthropic and OpenAI choose who carries out the reviews [13].
The dates sit close together. The Oct 5 position was eight days old when Reuters published on Oct 13 [23][21]. By Oct 8, three days after it, one person said Anthropic's talks with some evaluators appeared to have evolved [22][19]. We think that points to an opening ask that met resistance, though the sources do not say who pushed back, and a single anonymous person supplies the Oct 8 change. The people declined to be named because the matter was private [20].
Reuters tied the dispute to labs seeking trillion-dollar valuations in what could be major IPOs [14]. In our view that raises the cost of each redaction decision, because a risk finding in a public report is a document a prospective investor can read (this is our inference).
The counter-thesis is that this is routine contract haggling converging on the public line. Faculty's chief executive used that line himself: "Evaluators are only useful if people can trust what they report, so independence and transparency are critical," Marc Warner told Reuters [8]. We would be wrong if the signed terms let evaluators keep their conclusions and flag major redactions in their reports. The Oct 5 asks would then have been an opening position, and the public statements an accurate summary of the result.
What to watch
- Signed terms between Anthropic and Faculty on access and reporting, and whether Faculty can publish without the lab's approval.
- Whether evaluator reports begin to flag major redactions, as OpenAI said evaluators can.
- Whether lab observers and the bar on simultaneous embedding at two labs appear in final contracts.
Information, not investment advice. This story is general information about companies and markets, drawn from cited reporting. It is not a recommendation to buy, sell or hold any security; check the cited sources before acting.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+15
- Incentives75
- Confidence55
Perspective Coverage
4 publishers- Builder
- Builder 19%
- Operator
- Operator 44%
- Investor
- Investor 37%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Anthropic CEO Dario Amodei said last month that external evaluators should have the right to publish key findings "without editorial control."
- [2]
OpenAI agreed days later, saying assessors should have "editorial independence."
- [3]
The language AI labs already use in certain contracts with firms that evaluate their models can bar those firms from publishing confidential information in their reports without the companies' consent.
- [4]
The AI companies are worried that evaluator staff working inside the companies' walls as part of embedded evaluation could take proprietary information to rivals.
- [5]
Measures under discussion to prevent intellectual-property leaks include having lab employees observe embedded teams as they work.
- [6]
Measures under discussion also include preventing evaluators embedded at one lab from working at another simultaneously.
- [7]
Anthropic said last month that Faculty, Accenture's specialist AI business, would be its first embedded evaluator, but said details on what evaluators can access and how they report findings were still being worked out.
- [8]
"Evaluators are only useful if people can trust what they report, so independence and transparency are critical," Faculty CEO Marc Warner told Reuters.
- [9]
Technology companies typically bar the firms they work with from sharing such information with others without their consent.
- [10]
Anthropic said that seeking significant control over embedded evaluators' publications did not accurately characterize all of its discussions with independent evaluators, and that in current conversations with evaluators it previously worked with it had made clear the reviewers control their methods, evaluation criteria, conclusions and what to publish.
- [11]
For new evaluator firms, Anthropic said it reserved the right to comment on and approve reports as the reviewers learn how its models operate and establish a working relationship.
- [12]
OpenAI said it was not seeking approval rights over the full text of reports produced by evaluators, that its standard practice is to redact confidential information, and that evaluators can disclose whether major redactions were made.
- [13]
Experts have said independence of evaluators is a concern as Anthropic and OpenAI choose who gets to carry out the reviews.
- [14]
The discussions highlight the tension between a need to make the world aware of powerful and potentially dangerous AI and demands to safeguard trade secrets by AI companies seeking trillion-dollar valuations in what could be major IPOs.
- [15]
Under a bar on simultaneous embedding, an evaluator could serve 1 of the 2 labs at a time, or 50% of the pair.
- [16]
What labs consider crucial intellectual property can overlap with information relevant to judging a model's risk, which can lead to arguments and redactions, said a separate person with knowledge of past dealings with the labs.
- [17]
Anthropic and OpenAI have sought significant control over what outside evaluators can publish, concerned that findings could reveal trade secrets, according to two people familiar with the discussions.
- [18]
As recently as October 5, Anthropic and OpenAI were seeking strict editorial control over what the evaluators could publish, according to the two people with knowledge of the matter.
- [19]
By October 8, talks between Anthropic and some evaluators appeared to have evolved, according to one person familiar with the matter.
- [20]
The people familiar declined to be named because the matter was private.
- [21]
The Reuters report is dated October 13.
- [22]
Three days separate the Oct 5 position and the Oct 8 change reported by one person.
- [23]
Eight days separate the Oct 5 position and the Oct 13 Reuters report.
Sources
3 independent publishers whose own reporting we read for this story.
- channelnewsasia.comExclusive-Behind AI industry's call for independent reviews, a fight over what gets made public
1 article · October 13, 2026
- cryptobriefing.comAnthropic and OpenAI reportedly seek control over evaluator publications, raising trade secret concerns
2 articles · October 13, 2026
- pymnts.comAI Labs Having Second Thoughts on Independent Safety Evaluations
1 article · October 13, 2026
- thenextweb.comAI labs sought control over what outside evaluators publish, Reuters says
1 article · October 13, 2026
Topics and entities
Topics
- AI lab transparency
- Independent AI Audits
- AI Safety Evaluation
Entities
- OpenAI
- Redwood Research
- Reuters
- Dario Amodei
- METR
- Anthropic
- Accenture
- Rumman Chowdhury