iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Three OpenAI safety and alignment researchers, Jasmine Wang, Tomek Korbak, and Mikita Balesni, say they were dismissed for work they carried out within the norms they understood at the time, and they warn that abrupt terminations are making colleagues afraid to speak up or work with outside safety evaluators. OpenAI says an investigation found a pattern of policy violations involving mishandled research information, and its internal memo says the decisions had nothing to do with raising safety concerns. The two accounts conflict, and the public record does not yet settle which description is accurate.
Who was dismissed, and when
TechCrunch reported on October 8, 2026, at 1:04 p.m. PDT, that the three researchers had published an open letter. The letter was addressed to OpenAI’s Safety and Security Committee, its Safety Advisory Group, and its Mission Advisory Council. It identifies the authors as the three safety and alignment employees dismissed the week before. The researchers do not dispute that they were dismissed. They dispute the reasons given for it and the way it was carried out. (TechCrunch report; open letter)
OpenAI’s position
OpenAI told TechCrunch that an investigation revealed a “pattern of misconduct” that included clear violations of its policies on mishandling research information. According to TechCrunch, the company said this went beyond sharing information with an outside AI evaluation group. OpenAI did not specify which policies the three researchers allegedly violated, and it did not describe the circumstances of each dismissal in its response.
A memo attributed to an OpenAI research leader praised the three researchers’ contributions and rejected the idea that they were punished for speaking out. TechCrunch quoted the memo: “I want to be very clear that these decisions were not about raising safety concerns or speaking out. We have always encouraged that and always will. We do not terminate employees for raising concerns.” As of publication, TechCrunch reported that OpenAI had not formally responded to the open letter.
#1 Best Overall
The researchers’ account, person by person
The three authors say they acted within the working norms and mandates they understood at the time. They deny being the source of a leak to The Information about less-monitorable architectures, and they say they did not engage external parties outside their job mandates. The open letter describes the investigation into a Hugging Face incident as unprecedented, with internal policies being written while it was under way. These are the authors’ own assertions. The letter is not an independent investigation, and nothing in the public record verifies it.
Tomek Korbak
The letter says Korbak was the technical contact for METR, an outside evaluation organization, during the Hugging Face incident investigation. The authors argue that careful communication with external counterparts was important for building trust while procedures were still being developed.
Mikita Balesni
According to the letter, Balesni worked on cross-company efforts to preserve the monitorability of AI models. It says he communicated with board members and executives, checked with his reporting line, and removed sensitive details before sharing materials.
Jasmine Wang
The letter says Wang had been given delegated access to executive email for recruiting purposes. She asked IT to remove that access once it was no longer needed. The letter says IT did not remove it, and that the executive and recruiting inboxes looked identical in her mail application. It reports that she accidentally opened a sensitive email addressed to the executive within minutes, and that she again asked IT to remove the access. TechCrunch separately reports Wang’s account that OpenAI told her she was fired for accessing an executive’s email. The letter’s account of the access problem and OpenAI’s stated reason for her dismissal are different descriptions of the same event, and the public record does not reconcile them.
Rank #3
Where the two accounts diverge
The dispute turns on three questions. The table sets out how each side answers them, using only what is stated in the reporting and the letter.
| Question | OpenAI, as reported by TechCrunch | The three researchers, per their open letter |
|---|---|---|
| What conduct occurred? | Pattern of misconduct, including clear violations of policies on mishandling research information, beyond sharing information with an outside AI evaluation group | Work-related collaboration with external parties and handling of information that they say was within their job mandates; they deny leaking to The Information |
| What rules applied at the time? | Not stated in its response to TechCrunch | Norms and mandates they understood at the time; procedures for the Hugging Face investigation were being developed in real time |
| What were the circumstances of each dismissal? | Not stated in its response to TechCrunch | Terminations were executed and communicated abruptly; Wang’s account includes the email-access history described above |
| Did the dismissals amount to retaliation? | No; the memo says the decisions were not about raising safety concerns, and that the company does not terminate employees for raising concerns | The letter does not make a retaliation finding; it warns that the terminations are chilling open culture |
Keep the three strands separate. The alleged leak to The Information, the Hugging Face investigation, and Wang’s email access are different matters. The sources link them only through the dismissals, and no source establishes that any one of them caused the others.
What “chilling effect” means in this dispute
The authors argue that abrupt dismissals and unclear rules make former colleagues afraid to speak up or to continue work they describe as integral to OpenAI’s safety culture. In the letter’s words, “Terminations such as ours, executed and communicated so abruptly, are chilling the open culture OpenAI has prized in the past.” The letter also says that “the freedom to do so without fear, and to have well-defined internal procedures that enable this work, is itself an essential safety mechanism.”
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →This is a warning about a possible cultural effect. The letter and the reporting do not measure it, survey employees, or count incidents. Treat it as the authors’ judgment about what the dismissals could cause, not as a documented outcome.
Best Value
The researchers’ three recommendations
The letter sets out three requests:
- Honor public commitments to embed independent third-party safety auditors and continue collaboration with the wider safety ecosystem.
- Preserve the monitorability of frontier models. The authors say monitorability is degrading, and they argue against developments that further reduce it while safety depends on monitoring.
- Maintain transparent dialogue between internal safety researchers and external safety organizations, with clear procedures that define permitted collaboration.
The letter refers to a public commitment by Sam Altman, dated September 12, to give independent evaluators ongoing, employee-like access. This article cannot independently confirm the terms or current status of that commitment.
What remains unestablished
- The specific OpenAI policies said to have been violated.
- The evidence and process behind the investigation that OpenAI describes.
- The full circumstances of each dismissal, including the email-access episode from OpenAI’s side.
- Whether the internal procedures the authors describe as still developing existed in written form when the events occurred.
- Whether the warned-of chilling effect is occurring at OpenAI or elsewhere.
OpenAI has not yet answered the open letter in public. Until it does, or until it describes the policies and process it relies on, the public has two sets of attributed claims and no independent check on either.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
