Sources: Anthropic and OpenAI have pushed for substantial control over what external AI evaluators can publish, out of concern that trade secrets could leak
Anthropic and OpenAI have sought significant control over what outside evaluators can publish, concerned that findings could reveal trade secrets …
Context & Ripple Effects
OpenAI and Anthropic have already pressed Washington to restrict open-source AI models and to apply government model reviews across competitors. Their push to control evaluator disclosures extends that policy agenda from which models are accessible to which findings about them become public.
A September report said staff at both labs feared evaluators might compromise security, providing the internal rationale for tighter publication boundaries. The immediate tension is between independent scrutiny and protecting model information that the companies regard as commercially or security sensitive.
First-order effects
- External evaluators working with OpenAI and Anthropic face tighter lab influence over the publication of findings, limiting what independent testing can place in the public record.
- OpenAI and Anthropic gain greater ability to screen disclosures that they believe could expose trade secrets or model-security details.
Second-order effects
- Independent evaluators have less scope to use public reporting to compare frontier models, making outside evidence less complete for customers, researchers, and policymakers assessing the two labs.
- Government review systems that OpenAI and Anthropic have urged regulators to extend across rivals may depend more heavily on confidential reporting arrangements rather than public evaluator findings.
Third-order effects
- If labs broadly set the disclosure terms for external testing, AI evaluation may become a controlled-access assurance function rather than a source of fully public accountability.
- The pattern would concentrate authority over both model access and model observability in the frontier labs, sharpening the trade-off between proprietary security and independent oversight.
The trend: Frontier AI labs are seeking to pair formal oversight with tighter control over the information outsiders can inspect and publish.