OpenAI safety insider David Robinson calls company’s culture ‘broken’ after resignation

An OpenAI safety insider has called the company’s culture “broken” after leaving his position and speaking out in The Atlantic. David Robinson, described as having led the writing of safety reports that accompanied OpenAI’s major product launches, resigned this week and published an editorial in the magazine.

The statement is notable because it comes from someone whose work was positioned near the intersection of product promotion and risk communication. In AI organizations, safety teams often serve as a bridge between engineering progress and public accountability: they help explain what a system can and cannot do, what failure modes were observed, and what guardrails are in place. When a person responsible for that kind of documentation publicly characterizes the internal culture as broken, the claim invites attention not only to the individual episode, but also to the broader question of whether safety messaging and safety practice remain aligned.

Robinson acknowledged that his own situation is “something of a cliché”: an employee at a leading AI company who issues a dire warning while resigning from their job. The article also said that some cynicism is understandable when people suddenly come out with such warnings.

That self-aware framing matters. Resignation-linked warnings can sometimes be read as personal grievance, post-departure dramatization, or an attempt to settle internal disputes in public. At the same time, public criticism from former employees can be one of the few mechanisms available when internal channels are perceived as inadequate or when an employee believes the public may be receiving an overly reassuring account of risk. The value of such an intervention often depends less on the dramatic moment itself and more on the specificity of the concerns raised, the evidence provided, and the degree to which the person has had direct responsibility for the relevant processes.

Robinson’s role was tied to safety communications for OpenAI’s major model releases.

This connection is central to understanding the piece. Safety communications are not merely marketing copy; they shape how users, developers, regulators, and the public interpret the reliability of new systems. If internal culture is described as broken, a reader may reasonably ask whether warnings were softened, risks were underemphasized, or deadlines and commercial incentives crowded out candid assessment. Without additional detail, those questions remain open, but they illustrate why a departure from a safety communications role can carry outsized significance: it touches the boundary between technical capability and public trust.

The broader lesson is that AI safety depends on more than technical safeguards alone. It also depends on organizational habits: whether employees can raise concerns without career damage, whether reports reflect genuine uncertainty, and whether public explanations remain consistent with internal review. As model releases continue to attract attention, statements like this serve as reminders that transparency is not only a communication strategy but also a test of institutional health. When a former insider frames the problem as cultural, the discussion shifts from a single product launch to the conditions under which companies decide what to disclose and how to prioritize caution alongside progress.