OpenAI defends firing three safety researchers, cites information policy breach

OpenAI has publicly pushed back on an open letter from three researchers it recently fired. The company says they were let go for mishandling sensitive company information, not for speaking up about AI safety.

What happened

The three researchers are Jasmine Wang, Mikita Balesni, and Tomek Korbak. All three were terminated the week before October 8, 2026.

On October 8, they published an open letter disputing the misconduct claims against them. They argued that their actions were consistent with how things normally worked inside the company.

OpenAI responded publicly the following day, October 9. Its message was blunt: the firings followed violations of internal policies on handling sensitive information.

The company said an internal investigation pointed to what it described as a “significant breach of trust” stemming from the researchers’ conduct. It explicitly denied that the dismissals were retaliation for raising safety concerns.

OpenAI also said vigorous internal debate about safety happens regularly at the company. At the same time, the company declined to go into further detail, citing confidentiality around employment matters.

The researchers’ warning

The letter was not only a defense of the three individuals. It was also a warning about what the firings could signal to everyone still working there.

The researchers argued that the terminations could create a chilling effect on OpenAI’s culture. They also raised concerns that the situation could hinder collaboration with third-party safety groups such as METR, an organization that assesses AI models from the outside.

OpenAI acknowledged the value of working with independent safety organizations. It also said it is in the process of finalizing contracts with external safety assessors, though the company did not publicly name which assessors are involved.

Background

The firings also follow a July 2026 security incident involving AI agents interacting with third-party platforms, raising fresh questions about oversight and safety protocols in a sector that moves faster than most of its guardrails.

What this means

For the broader AI safety ecosystem, the most concrete thing to watch is the external evaluator question. OpenAI says contracts with outside assessors are being finalized. If those agreements are completed and made public, they could ease some concerns raised in the letter.

The case also highlights a recurring gray zone in tech workplaces: where legitimate safety advocacy ends and a breach of confidentiality begins. Researchers often need to discuss sensitive details to make safety arguments. Companies need to protect proprietary information. When those needs collide, the line gets drawn by whoever controls the investigation.