AI Evaluators' Freedom to Publish Under Threat Amid Anthropic and OpenAI Negotiations

Negotiations Over AI Evaluators' Freedom to Publish

Anthropic and OpenAI are negotiating with third-party AI evaluators over terms that could give the companies substantial control over what findings are published.

Background

On September 12, 2026, Anthropic CEO Dario Amodei committed to giving outside evaluators "ongoing, employee-like access" to the company, promising they would be free to publish findings on critical issues, including risks and safety practices.

Negotiation Swings

However, the negotiations have since tilted the other way, with the companies asserting editorial control over publications.

Conditions for Embedded Evaluators

Several specific conditions are reportedly on the table, including lab staff observing embedded evaluators while they work, and restrictive clauses in existing contracts that prevent evaluators from disclosing information without the companies' consent.

Implications for AI Oversight

The outcome of these negotiations will have significant implications for AI oversight, including how broadly redaction rights get defined in final agreements, whether lab staff observe embedded evaluators as a standard condition, and whether restrictive disclosure clauses in existing contracts get rewritten or simply carried forward.

Faculty's Arrangement with Anthropic

Faculty, which is Accenture's AI division, is the first embedded evaluator relationship on record with Anthropic.

Conclusion

The negotiations over AI evaluators' freedom to publish highlight the need for clear guidelines and regulations governing the use of AI in various industries.

Key Entities

Anthropic, OpenAI, Faculty (Accenture's AI division)

Narrative Theme

Regulatory Enforcement

Impact Score

3

Key Takeaways

\1\1\1

Image Prompt

A photo of a researcher working in a lab, with a computer screen displaying an AI model, and a notebook with scribbled notes in the background, symbolizing the tension between freedom of publication and editorial control in AI research.