---
id: "2026-10-13-reuters-labs-sought-control-evaluator-publications"
url: "https://postcutoff.com/e/2026-10-13-reuters-labs-sought-control-evaluator-publications/"
as_of: "2026-10-14T00:22:00+02:00"
date: "2026-10-13"
date_precision: day
category: policy-safety
importance: 3
confidence: medium
status: [Partly confirmed]
sources: 4
editor: human editor
human_review: null
version: "2026-10-13"
---

As of: 2026-10-14 00:22 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/e/2026-10-13-reuters-labs-sought-control-evaluator-publications/

# Reuters: Anthropic and OpenAI sought significant control over what outside evaluators can publish, citing trade secrets

A Reuters exclusive of Oct 13, 2026 (Aditya Soni) reports, citing two people familiar with the talks, that Anthropic and OpenAI have sought significant control over what outside safety evaluators may publish, out of concern that findings could reveal trade secrets. Both companies had said in September that evaluators should publish without editorial control. As recently as Oct 5 both were seeking strict editorial control, the sources said; by Oct 8 Anthropic's talks with some evaluators had moved. Anthropic says evaluators it already works with control their conclusions and what they publish, but that it keeps a right to comment on and approve reports by new evaluator firms. OpenAI says it does not seek approval of full reports and only redacts confidential information.

## Key facts
- Source: Reuters exclusive, Oct 13, 2026, 'Behind AI industry's call for independent reviews, a fight over what gets made public'; Techmeme credits Aditya Soni. Based on 'two people familiar with the discussions' plus a third person with knowledge of past dealings with the labs. Reuters says it is the first report of these negotiations
- The earlier promises (Reuters): Dario Amodei said in September that external evaluators should be able to publish key findings 'without editorial control'; OpenAI said days later that assessors should have 'editorial independence'
- The concern: staff of evaluators working inside the labs ('embedded evaluation') could carry proprietary information to rivals. Measures under discussion: lab employees observing embedded teams as they work, and barring evaluators embedded at one lab from working at another at the same time (one source)
- Timeline per the sources: 'as recently as October 5' both labs sought strict editorial control over publications; 'by October 8' talks between Anthropic and some evaluators 'appeared to have evolved'
- Anthropic's reply (Reuters): the description does not accurately characterize all of its discussions; with evaluators it has worked with before, the reviewers control their methods, evaluation criteria, conclusions and what to publish; for new evaluator firms it 'reserved the right to comment on and approve reports' while a working relationship is built
- OpenAI's reply (Reuters): it is not seeking approval rights over the full text of reports; its standard practice is to redact confidential information, and evaluators may disclose whether major redactions were made
- Existing contracts: language in certain lab contracts with evaluation firms can bar them from publishing confidential information without the company's consent, according to a contract Reuters saw. (The Next Web's rewrite states the opposite; we follow the Reuters text)
- On the record: Faculty CEO Marc Warner (Anthropic's first embedded evaluator, part of Accenture): 'Evaluators are only useful if people can trust what they report, so independence and transparency are critical.' Rumman Chowdhury (Humane Intelligence; former US science envoy for AI): 'Evaluators have this very privileged access, but whether explicitly or implicitly, they know that it comes at a cost'
- Precedent cited: OpenAI redacted nonpublic information from the METR and Redwood Research report on the Hugging Face incident; the investigation, planned as two days at OpenAI, took six days over three visits, with a fuller set of transcripts arriving on the last. METR and Redwood said nothing important to their conclusions was removed
- Context (Reuters): independent evaluation is one of four measures in the one-page White House Accord on Super Intelligence, which does not say how reviews should be done; evaluators' own 'minimum conditions' letter has more than 200 signatories, including Geoffrey Hinton

## What happened

In September both Anthropic and OpenAI promised that outside safety evaluators would decide for themselves what to publish.
Reuters reports that in the contract talks that followed, both companies asked for much more control. Their worry is that people
from evaluation firms who work inside a lab could take proprietary information to a competitor, and that a report on a model's
risks can contain information the lab treats as a trade secret. A person with knowledge of past dealings told Reuters that the two
categories overlap, which leads to disputes and redactions.

According to the two sources, the labs' position was still strict editorial control on Oct 5 and had begun to soften, at least in
Anthropic's talks with some evaluators, by Oct 8. The companies' statements to Reuters draw the line differently: Anthropic gives
established evaluators control of their publications but keeps approval rights over reports by firms new to its models; OpenAI
says it redacts confidential information and lets evaluators say when major redactions were made.

## Why it matters

Independent evaluation is the main safety mechanism the labs themselves proposed (Amodei's essay, the White House accord, the
planned standards body). None of those texts says who decides what an evaluator may publish. Reuters shows that this question is
being settled in private contracts, with the labs holding the access the evaluators need. The METR and Redwood report on the
Hugging Face incident, which OpenAI partly redacted, is the one public example so far.

## Sources and limits

The Reuters page itself is paywalled and was not read. This entry is based on the syndicated Reuters text published by ETCIO
(Economic Times), read in full, and checked against the Techmeme headline and the PYMNTS and The Next Web summaries. The article's
central claim rests on anonymous sources and both companies dispute parts of it, hence confidence: medium.

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 166 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 105 days after its cutoff
- Gemini 4 Argon (training cutoff March 2026): 196 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 135 days after its cutoff

## Sources

1. [Reuters (exclusive): Behind AI industry's call for independent reviews, a fight over what gets made public (paywalled)](https://www.reuters.com/world/behind-ai-industrys-call-independent-reviews-fight-over-what-gets-made-public-2026-10-13/) (reuters.com, press)
2. [Reuters via ETCIO (syndicated full text): Anthropic, OpenAI face scrutiny over independent AI evaluations](https://cio.economictimes.indiatimes.com/news/artificial-intelligence/anthropic-openai-face-scrutiny-over-independent-ai-evaluations/134915453) (cio.economictimes.indiatimes.com, press)
3. [PYMNTS: AI Labs Having Second Thoughts on Independent Safety Evaluations](https://www.pymnts.com/news/artificial-intelligence/2026/ai-labs-having-second-thoughts-on-independent-safety-evaluations/) (pymnts.com, press)
4. [The Next Web: AI labs sought control over what outside evaluators publish, Reuters says](https://thenextweb.com/news/anthropic-openai-ai-evaluators-publication-control) (thenextweb.com, press)

## Changes

- 2026-10-13 (filed): Created (evening)

## Related

- 2026-10-13: [UK MPs question OpenAI, Anthropic, Google DeepMind and Meta on AI safety](https://postcutoff.com/e/2026-10-13-uk-commons-bist-committee-ai-safety-hearing/index.md)
- 2026-10-01: [OpenAI fires three safety researchers who allegedly shared confidential information with an outside AI safety organization](https://postcutoff.com/e/2026-10-01-openai-parts-ways-three-safety-researchers/index.md)
- 2026-09-29: [Trump hosts AI CEOs at the White House](https://postcutoff.com/e/2026-09-29-white-house-ai-summit/index.md)
- 2026-09-24: [Google, OpenAI and Anthropic plan an industry safety standards body, the 'Standards Authority for Frontier AI', without government oversight](https://postcutoff.com/e/2026-09-24-labs-plan-frontier-ai-standards-body-safa/index.md)
- 2026-09-18: [Anthropic and Accenture commit $1B+ to embedded third-party evaluation](https://postcutoff.com/e/2026-09-18-anthropic-accenture-embedded-evaluation/index.md)
- 2026-09-12: [Dario Amodei publishes "We Must Pace the Frontier", calling for a deliberate slowdown](https://postcutoff.com/e/2026-09-12-dario-amodei-pace-the-frontier/index.md)
- 2026-08-26: [METR and Redwood publish the first independent investigation of a frontier-lab agent misalignment incident (OpenAI–Hugging Face)](https://postcutoff.com/e/2026-08-26-metr-redwood-hf-incident-investigation/index.md)
- People: [Dario Amodei](https://postcutoff.com/person/dario-amodei/), [Geoffrey Hinton](https://postcutoff.com/person/geoffrey-hinton/)