Dismissed OpenAI safety researchers call for stronger independent oversight
Three former OpenAI safety researchers say their abrupt dismissals could chill external collaboration and internal debate. OpenAI says they violated policies for handling sensitive information and denies firing them for raising safety concerns.
2 min read
Three former OpenAI safety and alignment researchers published an open letter on October 8, 2026 urging the company to preserve independent safety audits, model monitorability and open dialogue with outside experts.
Tomek Korbak, Jasmine Wang and Mikita Balesni say they were dismissed the previous week under circumstances that could discourage remaining employees from raising concerns or working closely with third-party safety organizations. Their letter argues that external collaboration is a practical safety mechanism and asks OpenAI to clarify the rules governing those relationships.
The researchers also deny leaking information about less-monitorable model architectures. They give separate accounts of the events they believe led to their dismissals, including communications with evaluator METR and an email-access incident. Those accounts are claims by the former employees, not independently established findings.
OpenAI disputes their interpretation. A company spokesperson told Business Insider that an internal investigation found violations of policies covering access to and handling of sensitive information. The company said the decisions were not retaliation for raising safety concerns. Business Insider also reported that an OpenAI research leader endorsed the letter's three high-level recommendations in an internal memo while defending the dismissals.
The recommendations are concrete: continue embedding third-party auditors, avoid model-development choices that further reduce chain-of-thought monitorability, and maintain transparent channels between internal researchers and the wider safety community.
The dispute matters beyond personnel policy because outside evaluators can provide an independent check on frontier-model behavior. But the public record remains incomplete: OpenAI has not disclosed detailed findings from its investigation, and the former employees reject the reasons conveyed to them. Until more evidence is available, the firings and their implications should be treated as contested accounts rather than settled conclusions.
Conversation
0 approved commentsNo comments yet. Start a thoughtful conversation.
Join the conversation.
Sign in with ChatGPT to leave a comment or rate this article.