OpenAI Safety Employee Resigns, Citing Broken Company Culture

What Happened

An OpenAI employee working on safety has resigned from the company, citing what they described as a broken internal culture around safety practices. The departure adds to a pattern of high-profile exits from OpenAI's safety and alignment functions that has drawn sustained scrutiny from researchers, regulators, and the broader technology industry.

Background

OpenAI, the San Francisco-based company behind the ChatGPT family of products and the GPT series of large language models, was founded in 2015 with an explicit mission centered on developing artificial general intelligence safely and for the benefit of humanity. The company restructured from a nonprofit to a capped-profit model and has since raised billions of dollars from investors including Microsoft, which has committed approximately $13 billion to the company.

Safety and alignment research, the work focused on ensuring AI systems behave as intended and do not cause unintended harm, has been a central pillar of OpenAI's stated mission since its founding. The company maintains a dedicated safety team and has published research on topics including reinforcement learning from human feedback, interpretability, and model evaluation.

The resignation follows a series of departures from OpenAI's safety function over the past year. Ilya Sutskever, a co-founder and the company's former chief scientist, departed in May 2024. Jan Leike, who led the company's alignment team, resigned shortly after, stating publicly that safety culture and processes had taken a back seat to product development at the company. Several other researchers from the superalignment team followed. OpenAI disbanded its superalignment team, which had been tasked with solving the technical problems of superintelligence safety, in mid-2024.

OpenAI has faced additional scrutiny over its governance structure. In 2023, the company's nonprofit board briefly removed chief executive Sam Altman before reinstating him days later following pressure from employees and investors. The episode prompted renewed debate about the effectiveness of OpenAI's governance model in holding management accountable to its safety mission.

What the Departure Adds to the Record

The latest resignation represents a continued pattern rather than an isolated event. Former employees who have spoken publicly after leaving the company have described tension between the pace of product releases and the time required to conduct thorough safety evaluations. OpenAI has disputed characterizations that it deprioritizes safety, stating publicly that safety remains central to its work and that it conducts extensive pre-deployment testing on all major models.

OpenAI has also committed to sharing safety information with other leading AI developers and has signed voluntary safety commitments with governments including the United States and the United Kingdom. The company has published system cards for its major model releases, documenting known risks and the mitigations applied before deployment.

The departure comes as OpenAI is in the process of completing a corporate restructuring that would convert its capped-profit subsidiary into a public benefit corporation, a change the company has said is intended to support its long-term mission while allowing it to raise capital at the scale required for frontier AI research. That restructuring remains subject to regulatory review in California and Delaware.

Industry and Regulatory Context

Safety staffing and culture at frontier AI laboratories has become a point of focus for policymakers in the United States and Europe. The U.S. AI Safety Institute, housed within the National Institute of Standards and Technology, has pursued agreements with leading developers to conduct pre-deployment evaluations of frontier models. The European Union's AI Act, which entered into force in 2024, includes provisions governing high-risk AI systems and general-purpose AI models with significant compute thresholds.

Competing laboratories including Anthropic, Google DeepMind, and Meta AI each maintain dedicated safety research functions, and the staffing and institutional independence of those teams has become a factor in how the companies are assessed by researchers and policymakers.

OpenAI's ongoing corporate restructuring is expected to face further review from California's attorney general in the coming months.