David Robinson, a leader on OpenAI's safety systems team best known for drafting the public transparency documents that accompany major model releases, has resigned from the company, Business Insider reported Friday.
Robinson's exit lands four days after OpenAI fired three safety researchers for allegedly mishandling confidential information, and it deepens a months-long pattern of departures and reorganizations across the company's safety apparatus — the very teams responsible for evaluating and disclosing how OpenAI's models could go wrong. For more context on this story, see our ongoing latest AI developments.
The Public Face of OpenAI's Safety Reporting
Robinson did not work on frontier training runs. His work was the paperwork the public actually sees: OpenAI's system cards, the documents released alongside each major model that describe what a model was tested for, where it falls short, and what safeguards apply. Startup Fortune described them as reading "like a nutrition label for AI" — a plain-language summary of testing and risk that regulators, enterprises and journalists rely on when evaluating a new release.
Writing on LinkedIn, Robinson said he served as lead drafter on Version 2 of OpenAI's Preparedness Framework — the company's methodology for assessing whether models pose catastrophic risks — and helped frame the company's disclosure that its Deep Research system was "on the cusp" of reaching High capability in biological-risk evaluations.
That disclosure was itself significant: under OpenAI's own framework, High capability in a domain like biology triggers formal security requirements and closer oversight. The person who put that assessment into public view has now left the company.
Robinson never held one of the company's flashy executive titles, but his output was arguably more visible to the outside world than that of most researchers: every system card a journalist, regulator or enterprise buyer read after a model launch passed through his drafting. As Startup Fortune noted in its report on the resignation, losing that role quietly is its own kind of signal about how much the company still values the function.
The Firings That Set the Table
Robinson's resignation surfaced days after OpenAI terminated three safety researchers — Jasmine Wang, Tomek Korbak and Mikita Balesni — for allegedly sharing confidential information, including details about the company's infrastructure architecture, with an outside AI safety organization. TechCrunch reported the firings on October 1, citing the Wall Street Journal. OpenAI has maintained that the terminations were about the handling of sensitive material, not about retaliation for raising safety concerns.
Whatever the full facts, the result is the same: four people connected to safety transparency and safety research have left within a single week, from a function that was already being reshaped. The BBC, which first reported on the terminations, described the affected workers as fired for "mishandling sensitive information."
A Safety Organization Repeatedly Reshuffled
The context stretches back further. Johannes Heidecke, who led OpenAI's safety systems group, departed in July 2026 after the company folded safety into its research division, according to reporting by Wired later picked up by Bloomberg. The merged team now reports to Mia Glaese, OpenAI's VP of Research and Safety, with Saachi Jain holding the safety systems lead role on an interim basis.
Earlier, the Financial Times reported that OpenAI had disbanded its dedicated Preparedness team — the unit built specifically to assess whether the company's own models could cause catastrophic harm. OpenAI disputed that characterization, saying preparedness work continues within Safety Systems. Joshua Achiam, a near-decade veteran of OpenAI's safety research who became the company's Chief Futurist, also announced a departure in the same stretch.
The upheaval is not happening in a vacuum. It is the same stretch in which Yann LeCun and Anthropic CEO Dario Amodei have publicly sparred over how seriously the industry takes catastrophic AI risk — a debate Startup Fortune noted as context for the departures. OpenAI keeps insisting, in system cards and interviews, that it takes the risk seriously. The people actually writing those disclosures keep leaving.
Why It Matters Ahead of a Possible IPO
The timing is awkward for more than optics. OpenAI has confidentially filed draft paperwork for an initial public offering, and while CEO Sam Altman has said the company will not go public in 2026, a listing would put the company's risk disclosures — exactly the kind of documentation Robinson specialized in — under intense legal and regulatory scrutiny from bankers, lawyers and securities regulators.
Critics of frontier AI labs have long argued that safety functions are the first casualty of commercial pressure. OpenAI says its safety commitments are unchanged and that preparedness evaluations continue. But the visible workforce behind those commitments keeps thinning, and for now, the people whose job was to tell the public what these models can do keep walking out the door.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
