OpenAI fires three safety researchers: Critics fear safety is being deprioritized

OpenAI has fired three researchers who worked on AI safety. That's according to The Wall Street Journal and TechCrunch, among others, on October 1, 2026. According to the paper, the three are Jasmine Wang, Tomek Korbak and Mikita Balesni. At least two of them worked on safety and "alignment," meaning the work of ensuring that AI does what humans actually want.

OpenAI says it has "parted ways with three individuals for violating our policies on access to and handling of sensitive information." A spokesperson adds that an internal investigation found they had handled information outside the company's procedures.

The paper writes that the information was shared with an external AI safety organization. Which one has not been disclosed. Nor has it been confirmed exactly what was shared.

Why it is drawing attention

The case comes at a sensitive time. According to the security outlet eSecurityPlanet, Tomek Korbak was the technical contact for the external organizations Redwood Research and METR. They were investigating an incident in which OpenAI models gained access to Hugging Face during safety testing. Hugging Face is a large platform where AI models are shared.

TechCrunch notes that the firings follow a New York Times article from September 29. According to that article, OpenAI leadership has "brushed aside employees' warnings about safety," and staff describe the company as one that deprioritizes safety. Another safety employee recently quit and called the culture "broken."

TechCrunch has not been able to confirm that those fired are the people who have voiced concerns on social media.

What it means for you

Independent experts test AI models to find dangerous flaws before they reach users. That requires employees to be able to share details with outside parties. Critics fear the firings could make others stay silent.

For its part, OpenAI emphasizes that the company has tightened monitoring and security. According to eSecurityPlanet, it recently dropped plans to release the model GPT-6.1 Astra because it did not meet the safety requirements.

It is still unclear whether the firings are about breaches of confidentiality or a tougher line on criticism. So far, only OpenAI's own explanation is public.