Three dismissed OpenAI safety researchers say they were never given a specific explanation of what confidential information they allegedly leaked. According to the reporting, they denied allegations involving discussions of reasoning monitoring with outside researchers, and an external collaborator also denied receiving secrets. OpenAI says the dismissals concerned policy violations, not the raising of safety concerns, and that use of unapproved Signal chats was not the sole reason. Public evidence does not settle the competing accounts. The broader issue is how clearly and consistently confidentiality rules are applied to outside verification of safety research.