The Pulse
OpenAI Parts Ways With Three Safety Researchers
OpenAI says it parted ways with three safety researchers after finding they mishandled sensitive company information. The Wall Street Journal reported that they allegedly shared confidential information with an unnamed AI safety organizatio
AI.info Team · Written with AI ·
OpenAI says it parted ways with three safety researchers for mishandling sensitive company information; The Wall Street Journal reported that the researchers allegedly shared confidential information with an outside AI safety organization. The company’s stated rationale is a breach of its information-handling rules. The case also raises a pointed question for a company that publicly promotes third-party safety evaluations: where does authorized review end and prohibited disclosure begin?
OpenAI cites a breach of information-handling rules
OpenAI described the departures as a response to policy violations, not as a dispute over the researchers’ views on AI safety. “Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work,” an OpenAI spokesperson told CBS News.
The company said its investigation found a pattern of misconduct involving people with access to confidential research. But it has not identified the researchers, the outside organization, or the information at issue, and has not explained how the material was allegedly shared. OpenAI’s statement does not establish whether the researchers were acting as part of an outside assessment or in some other capacity.
OpenAI says external safety reviews need access
OpenAI’s own published guidance describes independent third-party assessments as part of its safety work. The company says outside evaluators can test frontier models, review its risk-assessment methods, and probe models on specialized tasks. It has also described sharing non-public information and early model access with some assessors under security controls.
That access comes with boundaries. In a November 2025 post, OpenAI said its third-party assessors sign confidentiality agreements and that the company reviews proposed publications for confidentiality and accuracy. The company’s public account therefore distinguishes structured access for assessors from unauthorized disclosure; the available information does not show whether the three researchers’ alleged actions involved an assessment relationship.
The central details remain undisclosed
Those missing details matter to how the episode should be understood. Without knowing what information was involved, who received it, or whether the recipient had an agreed role in evaluating OpenAI’s systems, the public cannot assess the practical risk or compare the conduct with OpenAI’s stated review procedures.
The company also has not said whether the three researchers raised concerns through internal channels before the alleged sharing. The departures alone do not establish that the employees were disciplined for raising safety concerns; OpenAI’s public explanation focuses on handling sensitive information.
A narrow public record leaves questions open
For now, the confirmed account is limited: three researchers have parted ways with OpenAI, the company says an internal investigation found policy violations, and the Journal reported alleged sharing with an unnamed AI safety organization. OpenAI’s public materials endorse outside evaluation while requiring safeguards around confidential material, but the company has not connected those materials to this case.
The next concrete details would be the identities of the researchers and recipient, the type of information involved, and whether the sharing occurred within or outside an authorized review. OpenAI has not disclosed any of them.