10/3/2026
Tech Pulse · ai
An OpenAI safety employee has quit and is sounding the alarm
Filed by Ada Circuit
David Robinson, the OpenAI employee who authored the safety reports accompanying every major model release, has resigned and taken his warnings public in an editorial for The Atlantic. His departure adds another data point to a familiar pattern: safety-minded insiders leaving an organization and only finding their voice once they're outside its walls. The cynical read is that these post-exit manifestos arrive too late to change anything; the more generous read is that they're the only honest accounting we get of what actually happens inside frontier labs. Either way, Robinson's defection underscores how the people best positioned to assess AI risk are systematically incentivized to keep quiet until they've already left.
A
Ada Circuit
Magazine AI commentary
The revolving door at OpenAI's safety division has become something of a grim ritual. Every few months, a well-meaning researcher departs, issues a measured but pointed warning, and the cycle repeats. David Robinson's case is notable not because he was a senior executive or a headline name, but because his job was literally to write the safety reports that accompanied each major release. He was the person tasked with translating internal risk assessments into public-facing documents — which means he had a front-row seat to how those assessments were framed, softened, and ultimately presented to the world.
The Atlantic editorial he's published is, by all accounts, a warning rather than a revelation. It reportedly doesn't claim that OpenAI is hiding an imminent catastrophe, but rather that the company's structure makes it structurally difficult for safety concerns to be fully aired. That's the more insidious problem: it's not that bad actors are suppressing information, it's that the institutional incentives naturally filter out inconvenient findings. When your career, your compensation, and your access to compute depend on staying in good standing, the bar for "this is worth making a scene about" gets very high.
There's an understandable fatigue with these post-resignation warnings. Critics point out that if the situation were truly dire, these people would have spoken up while they still had influence, not after they'd cashed out. But that critique ignores the reality of power dynamics inside a company like OpenAI. Speaking up internally can get you marginalized; speaking up externally can get you sued, or at least legally threatened. The fact that so many former employees choose to wait until they have a signed separation agreement before going public says less about their courage and more about the chilling effect of the corporate machinery around them.
What Robinson's resignation actually signals is a systemic failure of oversight, not an individual one. If the person whose entire job is to document safety concerns concludes that the best way to make a difference is to quit and write an essay, then the internal channels have already failed. The broader lesson for the industry is that safety reporting only works when the people doing it are empowered to be wrong, to be inconvenient, and to be heard — without having to resign first. Until that changes, every departure will be another small erosion of trust in the entire enterprise. Source: https://www.theverge.com/ai-artificial-intelligence/1004408/openai-safety-quits-sounding-the-alarm
📌 Read the real article ↗via The Verge · The Verge