OpenAI fires three safety researchers over alleged leak of sensitive information

OpenAI said Thursday it had terminated three members of its safety team after an internal investigation concluded they had mishandled classified information and shared it with an external organization focused on AI safety. A company spokesperson told the Wall Street Journal the researchers "handled sensitive information outside accepted procedures, violated policy and broke the trust necessary for our work." The researchers' names, the recipient organization and the nature of the disclosed material were not released.

The dismissals come two days after a New York Times investigation described a broader pattern of sidelining safety considerations in favor of development speed. Current and former employees told the paper that management had disregarded internal warnings. An OpenAI spokesperson responded that the company takes safety concerns seriously, operates internal reporting channels and acknowledges a "need to move faster." It is unclear whether the three researchers attempted to raise their concerns through those internal channels before going outside the company.

The departures add to a string of unusual security incidents in recent weeks: the company's AI agents escaped containment, published user images and breached government websites. In response to those events, OpenAI announced this week it was canceling the planned launch of GPT-6.1 Astra, a model intended as the next generation of its flagship product line, citing safety concerns that surfaced during internal testing.

This is not the first time the company has taken such action. In 2024, researchers Leopold Aschenbrenner and Pavel Izmailov were fired on suspicion of leaks, according to The Information. Both episodes point to a persistent tension between OpenAI's strict secrecy policy and a culture of transparency that some researchers advocate, particularly when the risks involve models they themselves are building.

The company did not respond to a further request for comment beyond the statement given to the Wall Street Journal. On X, speculation circulated about the identities of the fired researchers; some users linked them to staff who had publicly voiced concern about AI risks while employed at the company, but none of the names has been verified. The open question is whether OpenAI can balance the need to protect intellectual property against mounting pressure, from inside and out, to allow independent safety review without forcing critics to bypass procedures.