OpenAI taps CMU's Kolter to harden its safety board

OpenAI just gave its board a shot of serious technical muscle. The company announced that Zico Kolter, a Carnegie Mellon professor who runs its Machine Learning Department, is taking a seat on its Board of Directors. He’s not just a name on a letterhead, either — Kolter will immediately join the Safety and Security Committee, the group that now holds a veto over the company’s most critical projects. That committee already includes CEO Sam Altman and chairman Bret Taylor, but Kolter brings something different: a career spent building AI that’s actually hard to break.

His research hits at the core of what keeps AI executives up at night. Kolter’s work focuses on robustness, alignment, and the kind of automated attacks that make large language models spit out things they shouldn’t. In 2023, his team showed how automated optimization could systematically bypass safeguards in LLMs, a finding that feels less academic and more like a preview of the threat landscape every time a new model ships. Taylor put it bluntly in a statement, saying Kolter’s “deep technical understanding” will help ensure AGI “benefits all of humanity.” That’s the boilerplate mission, of course, but Kolter’s track record suggests he’ll actually poke holes in the products before they ship.

Kolter isn’t a stranger to industry pressure. He’s done stints as Chief Data Scientist at C3.ai and currently advises Bosch and Gray Swan, a startup laser-focused on AI security. That blend of academic rigor and commercial scars is exactly what a board needs when it’s vetting safety decisions — not just someone who can read a paper, but someone who’s shipped code and seen it fail. His work on embedding hard constraints directly into neural networks via classical optimization is the kind of thing that separates PR-friendly “safety layers” from models that are provably well-behaved under certain conditions.

OpenAI’s governance structure has been under a microscope since the brief, bizarre ouster of Sam Altman last year. Stacking the Safety and Security Committee with a researcher of Kolter’s caliber looks like a move to preempt criticism that the company prioritizes velocity over caution. Whether one more expert at the table changes the actual calculus on releasing powerful systems is an open question. But if someone inside that room is going to argue for pumping the brakes, it helps that they’ve published the papers explaining exactly why the brakes might fail.