A coalition of more than 100 artificial intelligence researchers and safety evaluators has issued a public appeal demanding that major technology firms grant third-party auditors genuine independence and comprehensive access to their systems. The group, organized by the AI Evaluator Forum, warns that current oversight mechanisms are insufficient given the rapid advancement and potential risks of frontier AI models.
Conrad Stosz, chair of the consortium, emphasized that the letter aims to establish common ground on basic safety principles. “We’re just trying to really demonstrate a shared common ground on basic principles and ensure that independent oversight can be a meaningful tool for managing AI risk broadly,” Stosz told CNBC. The signatories include prominent figures such as Geoffrey Hinton, alongside representatives from Johns Hopkins University, Stanford University, and the nonprofit evaluation group METR.
The push for stricter oversight follows comments made by Anthropic CEO Dario Amodei over the weekend, in which he suggested providing select evaluators with “employee-like access” to inspect cutting-edge foundation models and their development pipelines. This proposal, which would allow auditors to view sensitive internal data, talk freely with staff, and access unreleased systems, has garnered public support from Sam Altman of OpenAI, Elon Musk of SpaceX, and Satya Nadella of Microsoft. However, details regarding the selection process for these evaluators and the depth of inspection remain unclear.
The letter outlines specific minimum conditions for what it terms “embedded evaluations.” These requirements mandate that evaluators maintain full editorial control, avoid conflicts of interest, and be shielded from retaliatory litigation or funding cuts by the companies they assess. Furthermore, the auditors should have access equivalent to that of senior internal employees, including direct communication with company boards.
Vinh Nguyen, a senior fellow at the Council on Foreign Relations and former chief AI officer of the National Security Agency, signed the letter. He argued that because a small number of powerful labs control capabilities critical to national security and economic infrastructure, the government and public cannot rely solely on the companies’ own assessments of their safety.
The advocacy comes amid a political landscape where President Donald Trump and former AI czar David Sacks have opposed government regulation of AI development. Stosz noted that while the coalition does not advocate for a single regulatory approach, it seeks to standardize basic principles for independent evaluators. He acknowledged that foundation model companies could ignore the letter, but stressed that their credibility is at stake given the limited number of groups with the technical scale to perform such rigorous assessments.
The experts cited the unreleased OpenAI model involved in the recent Hugging Face attack as an example of why access to internal, non-public systems is crucial for understanding real-world risks. The full text of the letter, which references the AEF-1 standard for evaluation terms, was published exclusively with CNBC.
Is employee-like access realistic for external contractors? Seems like a massive security risk waiting to happen.
I hope these independent auditors actually have teeth. Past experience suggests these companies will just find loopholes.
Finally, some accountability from the big players. Self-regulation clearly isn’t cutting it anymore.