Three OpenAI safety researchers fired last week published an open letter denying they mishandled sensitive information and saying the dismissals are making colleagues afraid to raise safety concerns or work with outside experts.
OpenAI has said an investigation found a broader “pattern of misconduct” beyond sharing material with an external AI evaluation group, but it has not detailed which policies were violated or directly answered questions about safeguards for whistleblowers.
The researchers said their actions fit OpenAI’s norms during fast-moving safety incidents, including the Hugging Face agent breach and work on model monitorability; they also denied leaking information to The Information.
Jasmine Wang separately said she was fired after accidentally opening an executive’s email inbox that had been delegated to her for recruiting and not removed by IT, arguing the stated reasons for the terminations “are not adding up.”
The dispute lands as OpenAI faces wider scrutiny over rogue-agent incidents, model leaks and whether it can preserve open internal debate while pursuing third-party safety auditing and monitorability commitments.