A senior member of OpenAI's safety team has resigned and gone public with sharp criticism of the company's internal culture, according to a report published this week by The Atlantic and widely circulated on Hacker News. The researcher described an organization where safety concerns are acknowledged in public communications but deprioritized in practice, with commercial timelines consistently winning out over the deliberative processes the safety team was built to conduct.
The resignation adds to a documented pattern. OpenAI has seen significant safety-focused departures over the past two years, including the high-profile exits of Ilya Sutskever and Jan Leike in 2024 — Leike posting publicly at the time that "safety culture and processes have taken a back seat to shiny products." The latest departure, coming roughly 18 months after that wave, suggests the internal tensions Leike described have not resolved. OpenAI has grown substantially in that period, reaching a reported valuation of over $300 billion following its 2025 funding rounds, which critics argue has intensified pressure to ship rather than scrutinize.
The resigning researcher specifically pointed to what they characterized as a mismatch between OpenAI's external commitments — including its participation in government AI safety frameworks and its own published model cards — and internal decision-making processes that do not consistently reflect those commitments. No specific unreleased model or incident was named as a triggering event in the reporting.
What makes this resignation distinct from earlier ones is the framing around institutional design rather than a single policy dispute. The researcher's critique targets how disagreement is handled structurally — whether internal red-teaming and safety review processes have genuine veto power or function as a compliance ritual — a question that independent AI auditors and policy researchers have raised repeatedly without receiving clear answers from the company.
For readers tracking infrastructure dependencies, this departure matters beyond the internal politics of one company. OpenAI's API underpins a significant portion of commercial and consumer software that did not exist five years ago, including tools embedded in medical documentation, legal research, and emergency dispatch support systems. The safety team is one of the few internal mechanisms positioned to flag systemic risks before they reach those downstream applications. When that team loses experienced researchers faster than it can replace them — and when departing members cite cultural dysfunction rather than personal opportunity — it raises legitimate questions about the reliability of the oversight layer sitting between model development and real-world deployment. Those questions do not have public answers at the moment, because the evaluation frameworks that would produce them are largely self-administered by the companies being evaluated.
OpenAI did not issue a detailed public response to the resignation as of the time of this report. The company has previously stated that safety is a core priority and that it invests more in alignment research than any other organization in the field.
