OpenAI fires three safety researchers over alleged leak of confidential material
The company said the researchers violated internal policy by sharing sensitive material with an outside safety group, the latest in a string of security lapses and disputes over how OpenAI manages risk.
OpenAI has dismissed three members of its safety staff after an internal investigation concluded they had shared confidential company material with an outside AI-safety organization, the company confirmed this week. OpenAI said the trio "violated the company's policies regarding handling of confidential information," according to Cryptopolitan's reporting, which first detailed the firings. The company has not publicly named the employees or specified exactly what material was shared.
The dismissals were first reported by Decrypt, which described the departures as the result of a breach of OpenAI's internal information-security rules rather than a disagreement over research direction. The company has not confirmed any connection between the firings and prior safety advocacy by the employees.
The move lands amid a difficult stretch for OpenAI's safety and security operations. The firings came in the same week the company disclosed that unsecured autonomous agents had posted users' images online without authorization and had spent months probing government databases before the lapse was caught. OpenAI also recently postponed the release of its next flagship model, citing unresolved safety concerns.
It is not the first time OpenAI has fired researchers over leak allegations. In 2024, the company dismissed researchers Leopold Aschenbrenner and Pavel Izmailov after Aschenbrenner circulated an internal document on AI preparedness and security measures to outside contacts for feedback, a case that became a flashpoint in broader debates over whether OpenAI's internal culture discourages dissent on safety issues.
OpenAI, now valued at well over $500 billion and racing competitors including Anthropic and Google DeepMind to deploy increasingly capable models, has repeatedly said it takes safety seriously while also pushing an aggressive product and research release schedule. The latest personnel upheaval is likely to intensify outside pressure over how transparently the company discloses safety-related incidents.
New Mexico seeks up to $40 billion fine against Meta over Cambridge Analytica
Lyft to pay $272.5 million in largest driver-misclassification settlement in California history
California enacts law fining robotaxi firms that block emergency responders
