The most watched company in artificial intelligence is facing an internal storm. Last week OpenAI dismissed three safety specialists, Jasmine Wang, Tomek Korbak and Mikita Balesni, who answered on Thursday with an open letter to the firm's safety oversight bodies contesting the decision. The letter argues that sensitive material was handled inside company rules, and warns that the firings have spread fear among colleagues. According to CNN, the trio describe the circumstances of their removal as suspicious and have taken the dispute public; this account is drawn from CNN reporting.
Management tells a starkly different story. The company says the three shared confidential material with an outside AI safety organization, breaking rules on accessing and moving sensitive data. In a lengthy public statement on October 9, leadership rejected retaliation claims and argued the violations went beyond sharing with evaluators. The full letter text obtained from Newsweek lays the gap between the two narratives bare, as the document published by Newsweek makes the conflicting accounts easy to compare.
The letter's core: a warning about fear
The sharpest passage concerns precedent. Conduct treated as routine a month ago is now a firing offense, the authors write, leaving staff unsure where they stand. They also allege that vague rumors are circulating internally to discredit them. Close teamwork with external experts, they argue, is indispensable to safety work, and the freedom to do it without fear counts as a safety mechanism in itself. This account comes from Cybernews reporting; according to Cybernews, the trio believe they were punished for pressing safety criticism.
The culture warning has echoed across American coverage. According to BusinessInsider, the letter puts the chilling impact of abrupt dismissals at its center, and that assessment gets wide space in BusinessInsider coverage. The arc from a workplace that once rewarded candor and open dissent to one where rules feel murky forms the backbone of the text. The claim that employees now hold back and that third-party scrutiny is weakening makes this a debate about management philosophy, not just employment law.
Background: tension building since 2024
The current crisis needs a two-year lens. In May 2024 the company's long-horizon risk unit, superalignment, was dissolved, and its leaders Ilya Sutskever and Jan Leike departed. Leike said on his way out that safety work had been eclipsed by flashy products, while the computing power once promised to the team stayed disputed. This history is drawn from Wired reporting; according to Wired, the unwinding of that unit is documented step by step, showing today's distrust has old roots.
Stakes are visible on the product calendar too. According to AlJazeera, the company shelved the planned release of its next-generation model GPT-6.1 Astra to cut risk while formal regulation is missing; that detail comes from AlJazeera reporting. Reporting obtained from IndiaToday stresses the trio's framing of choosing safety over near-term corporate interests, as IndiaToday coverage highlights. Industry-wide signals add weight, with leading AI firms reportedly probing numerous incidents around agent software.
What remains unverified, and what comes next
Honesty requires noting the gaps. The outside organization's name is undisclosed, exactly which data moved has no independent confirmation, and the company's long statement clashes with the letter at several points. Rather than declaring a winner today, the healthier stance is to watch how the oversight bodies respond, how talent flows shift, and how the external evaluation ecosystem absorbs this precedent. Where the line runs between outside scrutiny and trade secrets at frontier labs stands as the sector's most concrete open question.
| Item | Summary |
|---|---|
| Three names | Wang, Korbak and Balesni push back in open letter |
| Dispute | Management cites policy breach, trio alleges retaliation |
| Precedent | Oversight bodies' reply will set sector example |
AI commentary
"In my view the file's real value sits in the gray zone between the two stories rather than inside either claim. Both narratives are one-sided, so I am watching the boards' answer and the precedent it writes more than any verdict."
AI assessment
The company's counter-case deserves weight: guarding trade secrets, especially at a frontier lab, is a legitimate duty, and outside sharing needs boundaries. Treating every safety complaint as automatically valid could open a door to abuse; clear rules serve both sides.
Yet the documents at hand are one-sided narratives; neither the letter nor the corporate statement has faced independent audit. The nature of the shared data, what internal procedures truly permitted, and the rumor claims all await verification. Readers should treat this file as an ongoing inquiry, not a verdict.
Both camps have narrative incentives: the trio seek public backing by speaking to the press, while management aims to repair institutional trust with its long statement. Whatever the oversight bodies answer from here will tilt the balance and set an example for peer labs across the sector.
The practical takeaway is this: outside evaluation of AI safety survives only if experts can speak without fear and companies can still guard secrets. That balance is built with written, predictable procedures, not murky rules, and this episode shows why those procedures matter.
Sources
8 links; no other published story cites them. Stories sharing a link do not confirm each other; a source's origin is not inferred from how often it is cited.
openai · ai safety · open letter · company culture · outside scrutiny