OpenAI’s research leaders responded publicly on October 9 to a letter from three former safety researchers in a Newsroom post on X. The Associated Press and Reuters report that the company said an internal investigation found the researchers violated policies for handling sensitive information and described a “significant breach of trust” beyond the letter. OpenAI denied the dismissals were about raising safety concerns. The post also said contracts with third-party safety assessors were being finalized, with details to come in the coming weeks. The company did not identify the specific conduct or policy at issue.

The big change

  • What changed: OpenAI’s public answer now pairs its account of the firings with a stated plan to finalize third-party safety-assessor contracts and a public reaffirmation of model monitorability as an industry-wide commitment.
  • Why it matters: The researchers’ dispute concerns both their treatment and their work with outside safety groups. OpenAI’s commitment gives that second question a concrete next step, while the terms, access and scope of the assessor work remain unknown.
  • What to watch: OpenAI said details would follow in the coming weeks. Those details can show whether independent assessors receive the sustained access the researchers requested and how staff can share information with them while protecting confidential material.

This is a dated continuation of BIG CHANGE’s report on the researchers’ letter. In their original letter to OpenAI’s oversight bodies, they described three different situations. Tomek Korbak said he was told verbally that his dismissal related to how he communicated with METR, the nonprofit involved in assessing OpenAI’s Hugging Face incident. Mikita Balesni said his work with outside groups on monitorability was coordinated with his reporting line and that he removed sensitive details before sharing materials. Jasmine Wang said access to an executive’s email had been delegated for recruiting, that she asked IT to remove it, and that she reported an accidental click on a sensitive message. These are the researchers’ accounts, not independently established findings.

OpenAI’s post addresses two of the letter’s central requests. The full text reproduced by The Indian Express says the company was finalizing contracts with third-party safety assessors and remained committed to embedding external reviewers. It also says OpenAI agrees that preserving frontier-model monitorability requires an industry-wide commitment and that it continues to invest in the work. Those are company statements about plans and priorities; the post does not name the assessors, set out their access or publish contract terms.

On the dismissals, OpenAI said an investigation found information-handling policy violations and that its decisions were not about safety criticism. But the public post does not identify the information, the specific rule or the evidence supporting its conclusion. The Associated Press reported that OpenAI would not give further specifics about why the researchers were fired. Reuters reported that the company did not reveal the alleged violations. Neither publishes the investigation file or independently establishes which account is correct.

The employment dispute also does not resolve the separate technical debate about monitorability. The researchers say the ability to monitor frontier models is degrading; OpenAI says it agrees monitorability matters and is continuing to invest. The October 9 response reports no new measurements or evaluation results.

Sources & further reading