Another OpenAI Safety Leader Resigns — Sounding Alarm on AI Risks

David Robinson, a leader on OpenAI's Safety Systems team, has resigned. A company spokesperson confirmed the departure to Business Insider, marking another significant exit from the teams responsible for managing the risks of advanced AI. Robinson is not leaving quietly; The Verge reports he is now publicly “sounding the alarm” in an editorial after leaving his post.

This isn't just any researcher. Robinson was responsible for writing the safety reports that accompanied every major OpenAI model release. These documents are the primary public-facing evidence of the company’s safety diligence. His departure and subsequent criticism directly challenge the credibility of the very process he once managed.

A Pattern of Safety Exits

Robinson’s resignation is not an isolated event. It is the latest in a series of high-profile departures from OpenAI's safety-focused divisions. This trend points to a growing internal conflict between the pressure to ship increasingly powerful models and the mandate to ensure they are developed responsibly. When the individuals tasked with hitting the brakes all start heading for the exit, it suggests the accelerator is getting all the attention.

The context here is critical. The company recently dissolved its Superalignment team, which was dedicated to controlling future superintelligent systems, leading to the resignations of co-founder Ilya Sutskever and team lead Jan Leike. Leike publicly stated that “safety culture and processes have taken a backseat to shiny products.” Robinson’s exit reinforces that narrative. All sources confirm his role was squarely within the company's safety apparatus, making his departure a direct commentary on that function.

From Internal Process to Public Rebuke

Employees who feel their concerns are ignored internally often resort to public channels. The Verge notes that Robinson is taking his concerns to an editorial in The Atlantic, a move that escalates his critique from an internal matter to a public indictment. This pattern indicates that internal mechanisms for raising red flags may be insufficient or are being overridden by commercial priorities.

The Verge also highlights a potential cynicism one might feel about the recent wave of public warnings from former OpenAI employees. It’s a fair observation. However, the consistency of the message from multiple, credible sources is becoming difficult to dismiss as mere sour grapes. When the people who build the guardrails, study their long-term stability, and write the public reports on them all leave while citing similar concerns, it forms a coherent and worrying picture. It suggests a cultural shift inside one of the world's most consequential technology companies, where the race to build artificial general intelligence may be happening with fewer checks and balances than previously advertised.