What OpenAI’s Dispute With AI Safety Researchers Is About

Why did OpenAI fire the three researchers?

OpenAI says an internal investigation found that Jasmine Wang, Mikita Balesni and Tomek Korbak violated company rules for handling sensitive information. The company described the issue as a breach of trust and said the dismissals were not punishment for raising concerns about safety ¹².

The researchers give a different account of what the dismissals meant. In a letter to OpenAI leadership that they later shared publicly, they said they believed they were fired for putting safety ahead of the company’s short-term interests ¹. They also said they had acted within the working norms at the time ².

What safety concerns did the researchers raise?

The reports describe the researchers’ concerns broadly as concerns about AI safety. They do not set out a detailed list of the issues the researchers raised or tie a particular safety claim to the dismissal decision ¹².

The researchers’ public letter called for more transparency about why they were dismissed. They also said the firings could discourage other employees from speaking up, though the sources provided do not establish whether that effect has occurred ¹².

What does OpenAI say about the dispute?

OpenAI says the decision followed a thorough investigation into handling sensitive information. It has maintained that the reason was a policy violation, not the researchers’ views about AI safety or their decision to speak publicly ¹².

The distinction matters because the two accounts point to different explanations: a workplace rules issue, according to OpenAI, or a response to safety advocacy, according to the researchers. The reports do not include the investigation’s findings in enough detail to determine which explanation is better supported.

Why does this matter for AI safety oversight?

AI safety oversight includes the ways a company checks for risks and handles concerns raised by its employees. This dispute raises a practical question: can researchers raise difficult safety issues without fearing that their work or jobs will be affected?

That question is important beyond this case, but the sources do not show that OpenAI’s stated rules or investigation process are generally unfair. They also do not provide enough evidence to conclude that the dismissals were retaliation. For now, the public record described in the reports is a disagreement between the company and the researchers, with key details not disclosed ¹².

What is still unknown about the firings?

The reports do not identify the specific sensitive information at issue, explain exactly how it was handled, or provide the full evidence reviewed in OpenAI’s investigation ¹². They also do not give enough detail about the researchers’ safety concerns to compare those concerns with the company’s explanation.

Without those details, readers cannot independently assess whether the policy finding was justified or whether safety advocacy played a part. More information from the company, the researchers or an independent review could clarify the dispute.