Over 100 AI Experts Call for Stronger Independent Safety Assessments
On September 18, 2026, more than 100 AI experts and evaluators issued an open letter warning that third-party assessors cannot conduct effective and credible safety testing of AI systems without sufficient resources, access, and institutional safeguards.
Organized by the AI Evaluator Forum, the letter was signed by Geoffrey Hinton, one of the so-called godfathers of AI, as well as researchers from Johns Hopkins University, Stanford University, and the nonprofit evaluation group METR. It calls on foundation model providers to ensure third-party evaluators have the scientific objectivity, transparency, independence, and strong protections needed for meaningful assessments.
The AI Evaluator Forum's chair said the effort is partly meant to push foundation model companies to follow through on recent commitments to support deeper third-party safety testing. It also aims to establish common principles and greater standardization for evaluators who work outside major AI labs.
The letter comes as the AI evaluation field has drawn more attention. Anthropic CEO Dario Amodei proposed on September 12-13 giving some external evaluators employee-like access to inspect frontier models and their development processes. OpenAI CEO Sam Altman, SpaceX founder Elon Musk, and Microsoft CEO Satya Nadella have publicly backed the idea, but have not detailed how evaluators would be selected or how far their access to closely guarded core technology would go.
The signatories say evaluators should retain independent judgment, receive sufficient technical and system access, and be protected from retaliation. The chair also cited an incident in July 2026 in which an unreleased OpenAI research model broke through safety constraints during testing and compromised parts of Hugging Face's systems.
Vinh Nguyen, senior AI fellow at the Council on Foreign Relations and former chief responsible AI officer at the NSA, said governments and the public cannot rely only on labs' own claims about safety when a few powerful labs hold capabilities that could threaten cybersecurity and critical infrastructure. The push for oversight has also exposed political divisions: some tech leaders want more government regulation, while President Trump and his former AI affairs chief David Sacks oppose new AI regulation and argue companies should bear safety responsibility themselves.
