OpenAI says it fired 3 safety researchers over sensitive-information rules

In a statement reported by CBS News, OpenAI said: “We want to be very clear that these decisions were not about raising safety concerns or speaking out”. The company also said it encourages internal debate and works with independent safety assessors.

The researchers’ account centers on cooperation with outside groups. They said Balesni’s cross-company work concerned commitments to preserve AI monitoring, and that he removed sensitive details from materials before sharing them.

Balesni said he believed the researchers were fired for putting safety ahead of OpenAI’s near-term corporate interests. OpenAI disputes that the dismissals were about safety concerns, leaving the two accounts in sharp conflict over the reason for the decision.

What does OpenAI say the breach involved?

OpenAI’s stated reason is that the researchers violated policies for handling sensitive information. It described the conduct as a significant breach of trust that went beyond the account in the researchers’ letter.

That explanation draws a distinction between the substance of safety concerns and the handling of information. The researchers say Balesni’s work involved safety-related cooperation and that sensitive details were removed before materials were shared; OpenAI says its investigation found policy violations.

The dispute therefore turns on more than whether employees should be able to question a company’s safety decisions. It also concerns how staff and outside assessors can examine safety issues while following rules for sensitive information. OpenAI says it supports both internal debate and independent assessment.

What are the researchers warning about?

The three former researchers say the firings and communications about them could discourage colleagues from speaking openly about safety.

That is the researchers’ stated concern, not a claim that all OpenAI employees have stopped raising issues. Their warning focuses on how colleagues may interpret the dismissals and whether cooperation with outside safety organizations can continue without fear of retaliation.

OpenAI rejects the idea that it punished safety criticism. Its position is that the investigation concerned information-handling rules, while debate and work with independent assessors remain part of its approach to safety.

How do the firings test OpenAI’s safety commitments?

OpenAI’s stated support for internal debate and outside assessment sets a standard against which the dismissals will be judged. Enforcing rules for sensitive information and protecting good-faith safety discussion are both part of that challenge; one commitment cannot stand in for the other.

The company holds the power to investigate its employees’ conduct and make employment decisions. The researchers’ warning highlights the other side of that power: when safety staff see colleagues dismissed after work involving outside assessors, they may question how safe it is to raise concerns or share findings.

The researchers addressed their letter to OpenAI’s safety oversight groups and asked them to consider the effect on open debate. The central test for the company is whether it can enforce information-handling rules while keeping internal dissent and independent safety scrutiny genuinely welcome.