OpenAI Fires Three Safety Researchers Amid Dispute Over AI Risks

Computer-generated English audio

October 10, 2026 | United States. OpenAI’s dismissal of three safety researchers has triggered a public dispute over confidential information, independent AI oversight and the ability of employees to raise concerns about advanced systems. The company says the terminations followed violations of policies governing sensitive research information. The researchers dispute that explanation and warn that the dismissals could discourage critical safety discussions.

Who are the researchers?

The three former employees are Jasmine Wang, Tomek Korbak and Mikita Balesni. According to reporting by Reuters and the Associated Press on October 9, the researchers published a letter addressed to OpenAI’s safety oversight groups after their departures. Their letter challenged the company’s account of the events and argued that employees should be able to collaborate with independent evaluators while carrying out safety work.

What OpenAI says

OpenAI says an internal investigation found violations of clear policies for handling sensitive information and described the matter as a significant breach of trust. The company explicitly denies that it dismissed employees for raising safety concerns or speaking out. The publicly available statements do not establish all the underlying facts of the dispute, so the competing claims should not be treated as independently proven.

What the former employees say

The researchers say their work and communications were connected to legitimate safety responsibilities. Korbak has discussed his interactions with METR, an independent AI evaluation organization involved in reviewing an earlier AI-agent security incident. The researchers have also expressed concern that abrupt dismissals could have a chilling effect on other employees who want to question safety decisions or work with external oversight bodies.

Why outside AI evaluations matter

Independent testing can help identify weaknesses that an AI developer may miss in internal assessments. Such evaluations can examine the reliability, security and potentially dangerous capabilities of advanced systems. At the same time, companies handling proprietary models and sensitive infrastructure must control access to confidential research. The dispute illustrates the challenge of maintaining both meaningful external scrutiny and strong information-security procedures.

What remains unresolved?

Which specific information-handling rules were allegedly violated, and what evidence supports those allegations?
Whether the researchers’ communications were authorized as part of their responsibilities.
How independent evaluators will be able to access sensitive safety information in future.
Whether OpenAI’s internal processes can protect both confidential material and good-faith safety reporting.

The broader AI governance debate

The controversy comes as policymakers and technology companies debate how much AI safety should depend on voluntary company commitments rather than enforceable rules. Advocates of external oversight argue that third-party scrutiny can build public confidence; industry leaders also emphasize the importance of protecting sensitive technology and research. These goals are not necessarily incompatible, but they require transparent processes and clear boundaries.

What is independently confirmed and what remains disputed?

In October 9 reporting, Reuters and the Associated Press confirmed that OpenAI dismissed Jasmine Wang, Tomek Korbak and Mikita Balesni. The company says its investigation found breaches of rules governing sensitive information and maintains that the dismissals were not retaliation for raising safety concerns. The former researchers contest that explanation, describing their work with external evaluators as part of their safety responsibilities and warning that the firings could discourage scrutiny. OpenAI has not publicly disclosed the full details of the alleged violations. Those allegations, and the former employees’ explanation of the motives, remain disputed rather than independently established facts. Reuters, October 9; Associated Press, October 9.