OpenAI safety leader David Robinson quits, calls company culture 'broken'

David Robinson, the OpenAI safety leader who wrote the safety reports accompanying the ChatGPT developer's product releases, has quit the company, warning that its culture is broken and that AI firms are not "being nearly careful enough" about developing the technology.

Robinson explained his resignation in an essay published under the headline, "I quit OpenAI because its culture is broken." His departure adds a named, senior internal voice to a growing chorus of OpenAI insiders who have left the company with public warnings about how it handles the risks of the systems it builds.

Who is David Robinson and what did he do at OpenAI?

Robinson led the writing of the safety reports that accompanied OpenAI's product releases. That role put him at the center of the company's public accountability machinery: the documents that tell regulators, customers and researchers what a new model can do, where it fails, and what risks it carries.

When the person responsible for that reporting concludes the culture around it is broken, the claim carries weight beyond a routine resignation. It speaks directly to whether OpenAI's safety disclosures reflect genuine internal scrutiny or a process the company tolerates rather than prioritizes.

What exactly did Robinson say?

His essay's title states his core verdict plainly: "I quit OpenAI because its culture is broken." In the piece, he warned that AI companies broadly — not only OpenAI — were not "being nearly careful enough" about developing the technology.

Two claims stand out:

• The problem, in his assessment, is cultural. It concerns how the organization weighs safety against other priorities, not a single product failure.
• The warning is aimed at the industry. Robinson framed insufficient caution as a sector-wide condition, positioning his resignation as a signal to peers as much as to his former employer.

He joins other insiders who have previously urged the AI industry to take more care over the rapidly developing technology, a pattern of exits and public statements that has shadowed OpenAI through its rise.

Why does this resignation matter?

OpenAI is the developer of ChatGPT, the product that pulled AI into mainstream use and triggered the current investment and deployment race. The company's internal practices — how it tests models, documents risks, and decides what to release — have become a reference point for the entire industry.

Safety departures matter at OpenAI for a specific reason: the company has built its public position partly on the claim that it takes existential and societal AI risks seriously. Each high-profile exit that contradicts that claim erodes the trust that regulators, enterprise customers and the public extend to the company, and by extension to the sector.

Robinson's role sharpened that stakes. Safety reports are one of the few instruments outsiders can use to judge whether a release was handled responsibly. A leader of that function saying the underlying culture is broken suggests the documents may be only as strong as the environment that produces them.

Is this an isolated incident?

No. Robinson joins other insiders in urging the industry to take more care over rapidly developing AI. His essay fits into a longer sequence of departures and warnings from people who worked inside the company as it scaled, people who concluded that internal safeguards were not keeping pace with the technology or the commercial pressures around it.

That pattern is why a single resignation essay draws attention. It is not one disgruntled employee; it is another data point in a consistent trend of safety-focused staff leaving OpenAI and saying so publicly on the way out.

What should readers watch next?

The signal to track is whether OpenAI responds substantively — with changes to how safety work is staffed, resourced or reported — or treats the essay as another passing controversy. Robinson's public warning also puts pressure on other AI companies to demonstrate that their own safety cultures differ from what he describes, at a moment when the pace of model development shows no sign of slowing.