Over 100 specialists in artificial intelligence and evaluation have rallied around a clear point: without adequate resources and protections, independent AI safety reviews will come up short. They put that stance in a public letter released on Friday and provided it to CNBC ahead of wider circulation.
Conrad Stosz, who chairs the AI Evaluator Forum consortium behind the letter, said the aim is to show "shared common ground on basic principles" so independent oversight can actually help manage AI risk at large. Notable names on the letter include Geoffrey Hinton and people affiliated with Johns Hopkins University, Stanford University, and evaluator nonprofit METR.
The signatories want foundation model providers to make room for rigorous outside testing. Specifically, they call for third-party evaluators to have the "scientific objectivity, transparency, independence, and robust protections" required to do credible work. They also want evaluations to occur outside company control, with more openness about the systems being tested, and for evaluators "to be shielded from retaliation from the companies they embed with."
Stosz said the push is about holding model makers to their recent promises to support more thorough outside safety testing. The coalition is not backing a single method for safe development, he added, but wants basic standards and more consistency for independent evaluators and others working outside the big labs. The effort comes as AI faces sharper scrutiny, amplified by insider concerns over the risks tied to frontier models.
When uncertainty touches major industries, protecting your savings supports long term growth. Join Briefs Finance CEO Jaspreet Singh on September 29th for a FREE live investor workshop, How to Profit From A Dollar That's Losing its Value, where he shows how we're spotting investment opportunities as the dollar falls. Save your spot.
The small evaluator community drew fresh attention after Anthropic CEO Dario Amodei suggested giving select evaluators "employee-like access" to inspect cutting-edge models and their development processes. Stosz said the idea appears to go well beyond current access, potentially including use of company computers, candid conversations with staff, and the ability to "see sensitive internal data and unreleased systems."
"That type of access would give us much greater confidence and certainty about the actual risk, particularly for systems that they're using internally and not releasing," Stosz said, pointing to the unreleased OpenAI model implicated in the Hugging Face attack. OpenAI's Sam Altman, SpaceX's Elon Musk, along with Microsoft CEO Satya Nadella, have voiced support for Amodei's concept, but key logistics remain unanswered, including which evaluators would be chosen and how deep their inspections could run.
There is a bigger policy fight in the background: some tech leaders want government guardrails, while President Donald Trump, together with his onetime AI czar, David Sacks, have firmly opposed such regulation. As this plays out, watch where companies land on third-party access, how they operationalize protections for evaluators, and how transparent they become about systems under the hood. Those choices will shape trust, timelines, and spending across AI projects that may show up in the products you use and the funds you hold.
A careful approach to risk helps your portfolio weather change and prosper. Our CEO Jaspreet Singh is hosting a FREE live investor workshop, How to Profit From A Dollar That's Losing its Value, on September 29th. Sign up free to join him live.
