OpenAI is making a structural bet that independence will strengthen its safety culture. The company’s Safety and Security Committee (SSC) is spinning out into a fully independent board oversight body, chaired by Carnegie Mellon machine learning department director Zico Kolter. The committee now has explicit authority to delay model launches if safety concerns aren’t resolved — a concrete shift from advisory to enforcement power.
The committee’s roster signals seriousness. Alongside Kolter sits Quora CEO Adam D’Angelo, retired Army General Paul Nakasone, and former Sony general counsel Nicole Seligman. The group just wrapped a 90-day review of OpenAI’s safety processes, producing recommendations across five areas that the full board has adopted. Among them: independent governance, enhanced security, transparency, external collaboration, and a unified safety framework for model development.
On the security front, OpenAI plans to segment its internal information more aggressively, staff up around-the-clock security operations, and explore an industry-wide Information Sharing and Analysis Center for AI threat intelligence. The company also wants more external testing of its systems, pointing to collaborations with Los Alamos National Laboratory and unnamed third-party safety organizations as evidence it’s serious about outside scrutiny.
What’s unsaid here is whether this committee would have stopped anything controversial had it existed earlier. The press release frames the SSC’s review of the o1 model launch as a positive case study, but independent oversight is only as strong as the information it receives and the backbone it shows. The board will get regular briefings on technical assessments and post-release monitoring. Whether those briefings surface uncomfortable truths — and what happens if they do — is the real test.
