David Robinson didn’t mince words when he walked away from [OpenAI](https://openai.com). The safety researcher’s resignation letter, obtained by [TechCrunch](https://techcrunch.com/2026/10/03/openai-safety-employee-resigns-claiming-the-companys-culture-is-broken/), paints a picture of a company whose internal culture has fundamentally shifted as it chases the artificial general intelligence prize.
Robinson acknowledges he’s become “something of a cliché” – another AI safety employee issuing dire warnings while heading for the exit. But that self-awareness doesn’t diminish the weight of his concerns about how [OpenAI](https://openai.com) operates behind closed doors.
The timing couldn’t be more pointed. Robinson’s departure follows a string of high-profile exits from OpenAI’s safety teams, including co-founder Ilya Sutskever and safety researcher Jan Leike earlier this year. Both left citing concerns that the company was prioritizing flashy product releases over the careful safety work needed for increasingly powerful AI systems.
[OpenAI](https://openai.com) has been sprinting toward what CEO Sam Altman calls “superintelligence” – AI systems that could surpass human capabilities across virtually all domains. That race has intensified pressure inside the company, with multiple former employees describing a culture that’s shifted from cautious research to aggressive commercialization.
The pattern is becoming unmistakable. Safety researchers join [OpenAI](https://openai.com) drawn by the mission of developing beneficial AI, then leave frustrated by what they see as corner-cutting and cultural problems that make their work impossible. Robinson’s resignation adds another data point to this troubling trend.
What makes Robinson’s departure particularly significant is his willingness to go public with specific cultural criticisms. Previous exits have been more diplomatically handled, with departing researchers offering vague concerns about “alignment” between personal values and company direction. Robinson appears to be taking a more direct approach.
The safety team brain drain comes as [OpenAI](https://openai.com) faces mounting external pressure over its development practices. Competitors like [Anthropic](https://anthropic.com) have positioned themselves as the more safety-conscious alternative, while regulators worldwide are scrambling to understand and oversee AI development.
Industry observers note that losing safety talent isn’t just a PR problem for [OpenAI](https://openai.com) – it’s a technical one. These researchers possess deep knowledge of the company’s systems and the specific risks they pose. Their exodus leaves gaps in institutional knowledge precisely when that expertise is most needed.
The resignation also highlights the broader tension within AI companies between moving fast and staying safe. [OpenAI](https://openai.com) transformed from a research lab into a commercial powerhouse worth over $80 billion, but that metamorphosis has apparently cost the company some of its founding principles.
For Robinson, the cultural shift became impossible to ignore. His decision to speak publicly about a “broken” culture suggests the problems run deeper than typical workplace friction – pointing instead to fundamental disagreements about how AI development should proceed.
The question now is whether [OpenAI](https://openai.com) will address the underlying issues driving these departures or continue losing the safety expertise it claims to value. With each resignation, the company’s credibility on AI safety takes another hit, potentially undermining its position as regulators and competitors circle.
Robinson’s resignation crystallizes a growing crisis of confidence in OpenAI’s commitment to safe AI development. As the company races toward superintelligence, losing the very researchers tasked with ensuring that journey doesn’t end in catastrophe sends a troubling signal to investors, regulators, and the public. The real test will be whether OpenAI can rebuild trust with its safety teams before more critical expertise walks out the door.
