OpenAI confirms firing of three AI safety researchers after investigation

On Friday, OpenAI publicly dug in on its position in the case of three fired safety researchers: Jasmine Wang, Tomek Korbak and Mikita Balesni. The dispute has played out over two days. On Thursday, the researchers published an open letter demanding transparency. On Friday, the company replied on X that an internal investigation had found a "serious breach of trust" and that the decision stands.

What the company said

OpenAI's official wording rests on two points. First, the employees were fired for violating "clear rules for handling sensitive information." Second, the reason is not that the three spoke publicly about the company and voiced concerns about AI safety. The available portion of the material does not say what information OpenAI claims was mishandled. It also gives no details on how the investigation was conducted or who led it.

The X post reads as a direct reply to the letter. Its authors, as well as Wang, Korbak and Balesni in separate social media posts, asked the company to explain its decision in more detail. The available excerpt cuts off before the full list of their arguments, so summarizing it in full would be a stretch.

Key facts

• Three people were fired: Jasmine Wang, Tomek Korbak, Mikita Balesni.
• OpenAI's wording: a "serious breach of trust" found by the investigation.
• Stated grounds: violation of rules for handling sensitive information.
• The company's position: the firings are not connected to public statements about AI safety.
• The researchers' position: the decision must be explained publicly, as their Thursday letter says.
• Timeline: the letter was published on Thursday; OpenAI's response came on Friday, 9 October 2026.

Why this is a sore subject for OpenAI

The company already has a history of safety-related conflicts. In 2024, it disbanded the Superalignment team, and the departing Jan Leike said that safety culture had taken a back seat to shipping products. A scandal over nondisclosure agreements broke out around the same time. As a result, any dismissal of people who work on safety gets read through that lens, regardless of what the official statement says.

The dispute here is fundamentally asymmetric. The company says this is about discipline and protecting confidential data, and that is a legitimate argument: labs working on frontier models do keep model weights, evaluation results and internal reports on dangerous capabilities under lock and key. But OpenAI does not disclose the details of the violation, and no one outside the company can verify how serious it was or whether the punishment was proportionate. The researchers, for their part, cannot demand transparency and at the same time disclose what the company calls protected information.

In this situation, the two sides' arguments are hard to compare, and that is exactly what makes the conflict drag on. The reputational cost to OpenAI grows with each day, as long as there is no independent review or at least a more detailed explanation.

What to watch next

For anyone who works with OpenAI's products or is evaluating it as a partner, the practical takeaway is simple. Watch three things: whether the company discloses details of the violation, whether an independent assessment of the investigation emerges, and how other members of safety teams react, publicly or by leaving. If you are building an internal policy for handling sensitive data in your own AI team, this case is a useful example: the rules should be written down in advance, and the investigation procedure and firing criteria should be clear to employees before a conflict, not after it. For now, all anyone can work with is what is known: two public statements, diverging assessments and no verifiable details.