AI Labs Having Second Thoughts on Independent Safety Evaluations

By PYMNTS | October 13, 2026 | The call for oversight may be coming from inside the artificial intelligence industry’s own house, but the callers don’t want it going any further. Weeks after agreeing on a voluntary program to allow outside evaluators see inside their labs, OpenAI and Anthropic are seeking to control what the evaluators can do with the information they gather.

In a Tuesday (Oct. 13) Reuters report , sources said the AI companies are worried that evaluator staff working inside the companies’ walls as part of so-called embedded evaluation ‌could take proprietary information to rivals and want restrictions on what can be disclosed. Independent evaluation was one of four safety measures in the one-page agreement on oversight signed last month by President Trump and leading AI executives.

But the steps agreed to are voluntary, and the agreement did not specify how safety reviews ​should be conducted. Last month, both Anthropic and OpenAI said evaluators should have the right to publish key findings without editorial interference. But the labs have grown concerned that information relevant to judging a model’s risk could include sensitive intellectual property and other proprietary information that could be shared with rivals.

Measures now under consideration, per Reuters’ sources, include having lab employees observe embedded evaluation staff as they work, and barring evaluators working at one lab work at another. Anthropic and OpenAI are also reconsidering whether to impose editorial control over what evaluators could publish.

Anthropic told Reuters that in current conversations with evaluators worked with previously, it made clear that the outside reviewers control their methods, evaluation criteria, conclusions and what to publish. But for new ​evaluator firms, the company said it reserved the right to comment on and approve reports.

We’d love to be your preferred source for news . Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!

OpenAI told the news agency it was not seeking approval rights over the full text of ​reports produced by evaluators, but said its standard practice is to redact confidential information. Evaluators would be permitted to disclose whether major redactions were made. In an open letter signed last month by 200 researchers, including Geoffry Hinton, the potential evaluators called for independent control of their findings, broad access to AI labs and protection from retaliation.

The tension comes as policymakers and some tech executives ramp up calls for strict government oversight of AI companies rather than relying on voluntary self-policing by the industry. Several bills have recently been introduced in Congress aimed at requiring greater transparency by AI labs into their inner workings.

Last month, a trio of Democratic senators introduced the Artificial Intelligence Risk Management and Security Act, which would put the Commerce Department in charge of devising the technical standards for evaluating AI models’ risk and developing standard formats for publishing the technical details.

A separate bill, introduced by Republican Sen. Josh Hawley of Missouri, would put the Department of Energy in charge of stress-testing the most powerful AI systems before they reach the public. It would also impose severe financial penalties on companies that release new frontier models without going through prior review.

In an interview with the BBC on Monday, meanwhile, Microsoft President Brad Smith said the time for government action is now, citing recent incidences of labs losing control of their AI agents in the wild. “The moment has arrived,” he said. “Right now, your washing machine is subject to more safety rules than AI.

Every carton of milk that you see in the grocery store is subject to safety rules and inspection.” He called for global governments to agree on common rules for ensuring AI safety and requiring companies to disclose when rogue AI agents breach systems and to submit formal reports documenting those incidents.

For OpenAI and Anthropic, however, calls for greater transparency come as they seek trillion dollar valuations ahead of planned IPOs. The disclosure of valuable trade secrets could compromise those preparations by aiding competitors. It could also affect how potential investors scrutinize the companies and assess their risk.

With huge future debt obligations on and off their books, and a constant need for fresh cash infusions, the leading labs can ill afford any loss in investor confidence.