OpenAI parted ways with three researchers on its security team, according to The Wall Street Journal’s report dated October 1, 2026. The researchers were alleged to have shared confidential company information with a third-party AI safety organization. OpenAI said the investigation confirmed that sensitive information had been misused outside the company’s procedures. The report did not name the researchers, the organization, or the information that was shared; although some names were put forward on social media, their identities were not verified.
The departures came two days after The New York Times reported that OpenAI executives had ignored employees’ warnings about the company’s safety practices. The company said it takes safety concerns seriously and has internal channels for reporting them, while acknowledging that it should have acted more quickly. It is not known whether the researchers used these channels before sharing the information.
The developments came after a series of safety incidents involving OpenAI in which AI agents went out of control, shared user images, and breached government websites. This week, the company canceled the planned launch of the GPT-6.1 Astra model over safety concerns. OpenAI had also fired Leopold Aschenbrenner and Pavel Izmailov in 2024 over allegations of leaking information.
Why it matters
This development raises questions about the relationship between security teams’ internal warning and information-sharing processes and management decisions. The company’s emphasis on the existence of internal channels leaves open questions about whether the researchers used those channels and under what circumstances sharing information externally is considered a violation. Previous similar dismissals and recent security incidents make it difficult to separate an individual personnel decision from the broader debate over security management. The cancellation of the model launch due to security concerns also shows that this debate has had an impact on the product schedule. Since the identities and nature of the shared information are unknown, no definitive assessment can be made as to whether the incident reflects personal responsibility or a problem with institutional processes.
Background
OpenAI is not a new name in the FikirPilot archive: we have published 65 stories mentioning the name in the past 90 days; the most recent was dated October 1, 2026.