OpenAI and Anthropic are pushing for limits on what outside evaluators of their AI systems can publish, according to reporting. The issue centers on access to unreleased model capabilities for independent testing and how resulting findings are reviewed or redacted before going public.
BNN Bloomberg and Reuters-linked reporting say both companies want substantial control over publication because they fear that external evaluations could reveal trade secrets, even if the companies agree to allow independent scrutiny. The outlets describe a negotiation where researchers may be granted access to models, but publication rights would be subject to company approval and redaction rules.
The Medium piece frames the same tension more generally: independent researchers need transparency into unreleased systems, yet oversight by the companies can constrain what is ultimately disclosed. Across sources, the differing angles focus on the balance between openness and protection of proprietary information, and how that balance is being negotiated for third-party reviews.
