Over 100 artificial intelligence researchers and safety experts signed a public letter demanding that Anthropic and OpenAI establish independent oversight mechanisms for evaluating their foundation models. The letter, reported by CNBC, calls for transparency requirements and arms-length safety testing that separates evaluation teams from commercial product development.
The experts argue that current in-house safety evaluations lack credibility because the companies conducting them have direct financial incentives to accelerate product launches. Anthropic and OpenAI both perform internal safety testing before releasing new AI systems, but critics contend these evaluations operate without meaningful external accountability. The letter represents growing pressure on the two leading foundation model developers to adopt governance structures that remove conflicts of interest from the assessment process.
The signatories include academic researchers, AI safety specialists, and technologists from institutions such as MIT, Stanford, and UC Berkeley. They propose that independent evaluators should assess AI systems for risks including bias, toxicity, misinformation generation, and potential misuse before deployment. The letter stops short of naming specific model versions but addresses the industry-wide practice of companies self-certifying their own safety standards.
This pushback arrives as regulators worldwide examine how AI systems should be governed. The EU's AI Act already mandates risk assessments for high-risk systems. The UK's AI Bill of Rights and proposed U.S. frameworks similarly emphasize transparency and third-party verification. The letter taps into regulatory momentum, framing independent evaluation as both an ethical obligation and a practical safeguard against systemic risks.
Anthropic has publicly emphasized safety as a core differentiator, publishing research on constitutional AI and red-teaming practices. OpenAI has released safety documents and worked with external researchers on adversarial testing. However, both companies retain final control over which findings reach the public and how safety data informs product decisions. The letter suggests this arrangement remains inadequate.
The coalition does not call for halting AI development but rather for structural separation between companies that develop models and organizations that evaluate them. Independent evaluators would operate under transparent methodologies, publish findings with minimal editorial input from foundation model labs, and potentially face regulatory oversight themselves to ensure rigor.
The timing matters. Anthropic recently reported raising $5 billion in funding, while OpenAI's valuation exceeded $80 billion in secondary markets. Both companies face mounting pressure to ship new capabilities rapidly. Independent safety evaluation could slow product cycles, creating tension between safety advocates and commercial timelines. Companies may resist external mandates as regulatory burden, though the letter frames independence as ultimately beneficial for long-term trust and market legitimacy.
The letter signals that even as AI development accelerates, constituencies outside the leading labs are organizing to demand accountability. Whether Anthropic and OpenAI voluntarily adopt independent evaluation or whether regulators impose it remains an open question. The outcome will shape how foundation model development proceeds over the next 18 to 24 months.