David Robinson’s resignation landed amid an already bruising stretch for OpenAI’s safety operation. The former leader on its Safety Systems team, who worked on transparency and model system cards, left last week as scrutiny mounted over staffing, security incidents and whether the company was moving quickly enough to contain the risks of increasingly capable AI.¹

Robinson had spent three and a half years at OpenAI, overseeing parts of its preparedness framework and safety reporting. In his subsequent Atlantic essay, he described a workplace trapped in relentless motion: “my colleagues and I were so busy sprinting that we seldom had the chance to consider big changes.”² His conclusion was not that safety work inside the lab is pointless, but that it is insufficient without outside leverage. “Stronger incentives for safety — coming from outside the company — are a big part of getting this right,” he wrote.²

OpenAI offered a sharply different account of its posture. A spokesperson said the company was working to ensure that model capabilities do not outrun its controls, adding that it would “pause training or hold back models when we need to slow down.” The company also pointed to tighter security, third-party evaluation and earlier monitoring for concerning behavior.²

The dispute is larger than one resignation. Robinson joins a growing stream of safety-focused departures from leading AI firms, whose alumni argue that Silicon Valley’s “extreme confidence” and “perpetual sprints” are a poor fit for systems with potentially severe consequences. Their prescription is closer to nuclear plants or busy airports: redundancy, planning and humility rather than optimism alone.³

For Robinson, the next step remains unsettled. But his message is clear: the contest over AI safeguards may now be fought as much by public pressure and external institutions as by the companies building the models.²