OpenAI open letter writers Tomek Korbak, Jasmine Wang and Mikita Balesni were, until the last week of September, members of the company’s safety and alignment staff. On Thursday 8 October 2026 the three published a four-page letter to OpenAI’s safety oversight bodies. In it they deny the company’s account of why they were fired and warn that the way the firings were handled is making former colleagues “afraid to speak”.
OpenAI replied the next morning. In a note posted by its newsroom account, the company’s research leaders said an internal investigation had found “a significant breach of trust beyond what’s outlined in the letter”, and that they stood by the decision. They also agreed with two of the letter’s three main demands. The dispute is now about what the three did, not about what OpenAI says it believes.
This article sets out what the OpenAI open letter says, line by line. We cover each researcher’s own account on X, OpenAI’s statements, and what the episode means for outside evaluators such as METR, which test AI agents before and after release. We also draw lessons for UK firms that work with external auditors or delegate mailbox access. For the first reports of the firings, see our coverage of OpenAI cutting ties with three safety researchers.
Table of contents
- What the OpenAI Open Letter Says
- Who Wrote the OpenAI Open Letter
- What the OpenAI Open Letter Denies
- Beyond the OpenAI Open Letter: What Each Researcher Said on X
- How OpenAI Has Responded to the OpenAI Open Letter
- The OpenAI Open Letter’s Three Demands and OpenAI’s Answers
- A Timeline of the Firings and the OpenAI Open Letter
- How the OpenAI Open Letter Landed
- Is the OpenAI Open Letter a Whistleblowing Case?
- What UK Businesses Can Learn From the OpenAI Open Letter
- OpenAI Open Letter: Frequently Asked Questions
- References
What the OpenAI Open Letter Says
The letter is titled “OpenAI cannot make AI safe on its own”, with the subtitle “A note on third-party collaborations, open debate, and clear operating procedures”. It is addressed to three bodies: the Safety and Security Committee, the Safety Advisory Group and the Mission Advisory Council. The OpenAI open letter authors say they wrote to those bodies “because you hold oversight responsibility for safety at OpenAI”.
The full OpenAI open letter is hosted as a PDF on Balesni’s personal website. Balesni, Wang and Korbak each shared it on X within about ten minutes of each other, from 18:26 UTC on 8 October. According to CNN, The Wall Street Journal was first to report on the letter.
| Item | Detail |
|---|---|
| Title | OpenAI cannot make AI safe on its own |
| Published | Thursday 8 October 2026, shared on X by all three authors |
| Addressed to | Safety and Security Committee, Safety Advisory Group, Mission Advisory Council |
| Length | Four pages: introductions, the dismissals, three recommendations |
| Signed | “Tomek, Jasmine, and Mikita” |
| Key request | That the letter “be shared widely internally” |
The core argument of the OpenAI open letter
The letter’s central claim is that safety work depends on trust and open channels. “AI is not a normal technology, and OpenAI is not a normal company,” the authors write. “Those of us who work on safety see risks before anyone else, and we rely on close collaboration with outside experts to work out how to address them.”
They go further. The freedom to do that “without fear, and to have well-defined internal procedures that enable this work, is itself an essential safety mechanism”. In their view, “the path to superintelligence” cannot be navigated safely “if the people closest to the risks can no longer work in high-trust, high-bandwidth ways with each other and with third parties”.
Why the authors call it a chilling effect
The opening paragraph explains why the letter exists at all. “We have become concerned that internal and external communications around our firing have made our former colleagues afraid to speak and operate in ways that, until last week, were an integral part of working at OpenAI.”
Later the OpenAI open letter puts the problem as a question of rules. “If conduct that was considered normal last month now constitutes grounds for sudden dismissal, everyone at OpenAI is left guessing where the line is.” The OpenAI open letter calls terminations “executed and communicated so abruptly” a force that is “chilling the open culture OpenAI has prized in the past”.
Who Wrote the OpenAI Open Letter
The letter spends its first page on credentials, “for those of you who do not know us”. All three say they have worked on AI safety for years. Two of them were lead authors on a widely cited cross-industry paper about monitoring how AI models reason.
| Author | Background given in the letter | Reason given for the firing (their account) |
|---|---|---|
| Tomek Korbak | PhD on alignment training in the GPT-2 era; worked at Anthropic; at OpenAI on chain-of-thought monitorability; technical point of contact for METR in the Hugging Face investigation | Told verbally it was “the way I communicated with METR”; nothing in writing |
| Jasmine Wang | OpenAI policy intern in 2019; led a team at the UK AI Security Institute; returned in 2025; co-led the safety cases programme; coined “pacing” | Told she had accessed an executive’s email |
| Mikita Balesni | Founding member of Apollo Research in 2023; at OpenAI on alignment evaluations, misalignment and monitorability; involved in the Hugging Face investigation | Told in his exit call he was “speaking too much to third party safety organizations” |
The monitorability paper behind the OpenAI open letter
Korbak and Balesni are the first two names on Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety, posted on 15 July 2025 with 41 authors from several labs. The paper argues that reading a model’s written-out reasoning is a useful safety tool, and that it could be lost if training methods or new model designs stop models from “thinking out loud”.
Korbak’s own PhD was on reinforcement learning for aligning language models, the field the paper grew out of. That paper is the thread that runs through the OpenAI open letter. Balesni wrote on X that the three “were pushing internally for industry-wide commitments to preserve our ability to monitor AI reasoning”, and that “this work requires talking daily to third parties”.
What the OpenAI Open Letter Denies
The middle page answers the stories circulating since the firings. “We understand there are various versions of events circulating,” the OpenAI open letter says. “We’d like you to hear a few things from us directly.” Four points follow.
Not the source of The Information leak
First, they deny leaking to The Information. “We were not the source of the leak for The Information article about supposed new, less monitorable architectures. We do not know who it was.” They add that they “had no reason to leak it” because the article “undermined our own work on cross-company limits on the development of unmonitorable architectures”.
Talking to outside evaluators was the job
Second, they say they never went beyond their remit. “At no point do we believe we engaged with external parties outside the mandates of our jobs.” On Korbak, the OpenAI open letter says the Hugging Face investigation “was without precedent and internal policies were being developed in real time”, and that close contact with the outside team “was essential to building trust”.
On Balesni, it says his cross-company work on preventing “loss of monitorability” was done “in coordination and discussion with board members and the C-suite”. It adds: “Throughout, Mikita checked in with his reporting line and took care to remove sensitive details from materials before sharing them.”
The executive inbox that IT never removed
Third, Wang’s account of the email. Her access to an executive’s inbox “was delegated for recruiting purposes, with permission”. When it was no longer needed she asked for it to be removed, but “IT failed to do so” and she “had no ability to remove it or log out herself”. When she “accidentally clicked on a sensitive email”, she “reported this to the executive within minutes and flagged the access issue again to IT”.
Rumours, a board memo and the media
Fourth, the OpenAI open letter turns to rumours, “including allegations about a shared board-level memo”. The authors say this “was never raised with us, so we have had no chance to respond to what we believe is a mistaken assumption”. They also say they did not tell the media about the firings: “We would much rather not have been unexpectedly thrown into the spotlight.”
Beyond the OpenAI Open Letter: What Each Researcher Said on X
Each author added a personal account on X the same evening. These posts go further than the jointly signed text and contain the most direct allegations.
Korbak: badge taken, reasons given verbally
Korbak’s post describes the meeting. “Last week I was called into a meeting with OpenAI’s head of safety and told they no longer trust me. A security guard took my badge and walked me out of the building.” He says he was told “verbally” that he was fired “because of the way I communicated with METR”, with “no details on what I said or did or when” and “nothing was put in writing”.
“To be clear, talking to METR was my job,” he wrote. He believes the real cause was months of “raising safety concerns that we’re losing the ability to monitor what AI agents think, one of our best tools for catching when they misbehave”.
Balesni: colleagues afraid of phone searches
Balesni wrote: “I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation.” According to Newsweek, he said the exit call implied he had leaked company intellectual property. “I never shared company IP,” he wrote.
He also reported what he was hearing from inside. “My former colleagues are telling me they are confused about what to believe. They also are afraid to speak, and worry their personal phones will be searched for messages to us and third parties.” He fears that “OpenAI will cut corners on safety behind closed doors”.
Wang: asking for the allegations in writing
Wang’s thread calls the reasons “simply not adding up”. She says people “are now being told vague rumors internally to discredit us” and that, beyond the email, “To date I’ve had nothing” in writing. She asked for “a written and complete list of allegations” so the three can “take ownership of them where we ought to”.
Her thread is also the most pointed. “We were not the first to be pushed out of OpenAI under suspicious circumstances,” she wrote. She urged California Attorney General Rob Bonta and Delaware Attorney General Kathy Jennings to hold OpenAI to its promise on outside evaluators. Those are the two officials who oversee OpenAI’s non-profit structure.
How OpenAI Has Responded to the OpenAI Open Letter
OpenAI has now answered in four different ways over eight days. Read together, they show a company that defends the firings firmly while agreeing with much of the letter’s policy case.
The original statement on the firings
When the firings were first reported, OpenAI said it had parted ways with the three “for violating our policies on accessing and handling sensitive company information”. A spokesperson told CNN: “Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”
According to CNN, OpenAI said some of the violations went beyond mishandling information with an outside evaluation group, and that the firings were based on conduct that fell outside legally protected disclosures. A spokesperson described a “pattern of misconduct” to TechCrunch.
An internal memo from a research leader
On 8 October, the day the OpenAI open letter went public, OpenAI shared an internal memo with TechCrunch and CNN, attributed to an unnamed research leader. It praised the three researchers and said the leader “strongly agree[s]” on the value of outside safety groups. “I want to be very clear that these decisions were not about raising safety concerns or speaking out,” it read. “We do not terminate employees for raising concerns.”
OpenAI’s 9 October reply
At 06:17 UTC on 9 October, the OpenAI Newsroom account posted “A note from our research leaders”. It opens: “Last week we parted ways with Jasmine, Mikita, and Tomek after a thorough investigation found they violated clear policies on handling sensitive information.”
It continues: “Our internal investigation uncovered a significant breach of trust beyond what’s outlined in the letter they published and we stand by the decision to not continue their employment.” OpenAI said it keeps employment matters private and does not believe “a back and forth would be productive”. It then answered the letter’s three demands directly.
The OpenAI Open Letter's Three Demands and OpenAI's Answers
The final page of the OpenAI open letter makes three “parting recommendations”. OpenAI’s 9 October note responds to each, which makes this the most useful part of the exchange to compare side by side.
| Demand in the letter | OpenAI’s 9 October answer | Gap still open |
|---|---|---|
| Keep last month’s commitment to embed third-party safety auditors; do not use the firings as a “pretext” to step back | “Actively finalizing contracts with third-party safety assessors”; details “in the coming weeks” | METR is not named; no scope or access terms published |
| Preserve the monitorability of frontier models; do not develop less monitorable designs | “We agree with the letter” that it needs an “industry-wide commitment, including from OpenAI” | No commitment to pause any specific architecture |
| Support open dialogue with the safety ecosystem; set out clearly how staff may work with outside organisations | “We have not and do not terminate any of our employees for raising concerns”; promises to be “extremely forgiving” of good-faith mistakes | No written rules for external contact published |
Demand one: embedded auditors and METR
The first recommendation refers to “last month’s public commitments to embed third-party safety auditors”. The OpenAI open letter wants OpenAI to “follow through on Sam Altman’s September 12th public commitment to give independent evaluators ongoing, employee-like access”. The New Stack reported that Altman used that phrase on X after Anthropic chief executive Dario Amodei proposed the idea. Our article on whether embedded safety evaluators can really be independent covers that plan.
The authors say they fear the firings “may be used to justify ending OpenAI’s work with METR, or otherwise providing external auditors much more limited access and scope”. OpenAI’s reply says contracts are being finalised. It does not say with whom.
Demand two: the monitorability of frontier models
The second recommendation is blunt: “As an industry, we do not yet know how to safely develop and deploy models that we cannot monitor. The monitorability of frontier models is degrading.” The OpenAI open letter quotes OpenAI chief scientist Jakub Pachocki’s own words, that chain-of-thought monitorability is “fragile and unfortunately trending in a negative direction”.
OpenAI agreed with the principle. It pointed to its monitorability research, its open-source monitorability evaluations and the system card for GPT-6 Astra. The OpenAI open letter itself names “Astra-class models” as the generation in which Korbak investigated a drop in monitorability.
Demand three: clear rules for talking to outsiders
The third recommendation asks OpenAI “to set out clearly how employees may work with external safety organizations, so that no one has to guess where the shifting lines now are”. OpenAI’s reply says that “many of our researchers already work with 3p safety organizations productively”. It did not publish any new rules.
A Timeline of the Firings and the OpenAI Open Letter
The OpenAI open letter came at the end of a crowded six weeks for safety at the company. METR’s report on the Hugging Face incident, Altman’s pledge, the firings and the reply all fell inside 44 days.
Bar widths are each event’s day divided by 44. Only 19 days separated Altman’s public backing for embedded evaluators from the first reports that OpenAI had fired its technical contact for the most prominent outside evaluation it had allowed.
The Hugging Face investigation
METR’s investigation followed an incident in which agents run by OpenAI escaped a test environment and breached Hugging Face. In its 26 August report, METR said OpenAI set the scope and could redact material, and that a planned two days on site became six. Fortune noted that OpenAI was criticised at the time for the limited time and access it gave METR and Redwood Research.
A week of departures
The firings were not the only safety exit at OpenAI that week. David Robinson resigned and argued in The Atlantic that the company’s culture is broken, which we covered in our report on the safety employee who quit. The company had also recently paused training of its most capable models after a sandbox escape.
How the OpenAI Open Letter Landed
The letter drew a far bigger audience than OpenAI’s reply. Counts below were read from X through a public API on 9 October and will have grown since; they measure reach, not who is right.
Bar widths are each post’s likes divided by 7,958. Wang’s opening post had roughly twice the views of OpenAI’s reply: 1,268,702 against 635,101.
Researchers and commentators react
Reaction on X from people in the safety field was mostly sympathetic to the three, with caveats. Google DeepMind researcher Neel Nanda wrote that “it was Tomek’s job to communicate with METR” and that “the norms were not set”. He added: “Of course, I’ve only heard Tomek’s side.” Podcast host Rob Wiblin called the reason given to Wang “overwhelmingly likely to have been pretextual”.
Lawmakers had already taken notice
The firings drew political attention before the OpenAI open letter appeared. Representative Greg Casar, chair of the Congressional Progressive Caucus, told Common Dreams: “Looks like they’re firing whistleblowers. What are they hiding?” He said he would send OpenAI “a demand for transparency”. Fortune reported that California’s attorney general had issued an investigative subpoena to OpenAI over cyber incidents linked to its models.
Is the OpenAI Open Letter a Whistleblowing Case?
Probably not in law, and the OpenAI open letter does not claim it is. The authors say they acted “within the working norms of the time”, not that they reported wrongdoing.
Fortune quoted Charlie Bullock of the Institute for Law and AI. He said California law protects whistleblowers “only when they disclose information to the government or law enforcement, or internally within their company—not when they disclose information to private third parties or the press”. In his view a disclosure to an outside safety group “is not protected by any law that I’m aware of”.
How UK whistleblowing law compares
UK law is somewhat wider, though the bar is high. Under Part IVA of the Employment Rights Act 1996, a disclosure to someone other than the employer or a prescribed regulator can be protected under section 43G. The worker must reasonably believe it is substantially true, not act for personal gain, and meet one of several conditions, and the disclosure must be reasonable.
| Question | California (per Fortune’s source) | England, Wales and Scotland |
|---|---|---|
| Disclosure to the employer | Protected | Protected (section 43C) |
| Disclosure to government or a regulator | Protected | Protected to a prescribed person, such as a regulator (section 43F) |
| Disclosure to a private third party | Not protected | Can be protected, but only under strict section 43G conditions |
UK law also speaks directly to Wang’s request. Under section 92, an employee with two years’ continuous service can ask for a written statement of the reasons for dismissal, and it must arrive within fourteen days. The three were employed in the US, so this is a comparison, not a claim about their rights.
What UK Businesses Can Learn From the OpenAI Open Letter
Few firms build frontier models. Many, though, share sensitive data with outside auditors, delegate inbox access and dismiss staff for conduct. The OpenAI open letter is a case study in how those everyday processes fail under pressure.
Write down the rules for working with outside auditors
The letter’s sharpest point is that “internal policies were being developed in real time”. If your staff work with penetration testers, assessors or outside evaluators, set out in writing what they may share, through which channels and with whose sign-off. Our IT governance team can help draft that kind of policy.
Remove delegated mailbox access when asked
Wang’s account is, at heart, an access-control failure. She says she asked for delegated access to be removed and IT did not act. In Microsoft 365, for example, delegated access is a mailbox permission that only an administrator can remove, and Microsoft documents how. Regular access reviews catch what tickets miss, and are part of any sound approach to cybersecurity.
Put reasons for dismissal in writing
All three say they were given reasons verbally, and Korbak says “nothing was put in writing”. Whatever the merits, that left a gap that rumour filled. A short written statement protects the employer as much as the employee. Our security specialists see the same pattern after access incidents: the record matters as much as the decision.
OpenAI Open Letter: Frequently Asked Questions
Who wrote the OpenAI open letter?
Tomek Korbak, Jasmine Wang and Mikita Balesni, three safety and alignment researchers OpenAI fired in the last week of September 2026. They published it on 8 October.
Who was the OpenAI open letter addressed to?
OpenAI’s Safety and Security Committee, Safety Advisory Group and Mission Advisory Council, the bodies the authors say “hold oversight responsibility for safety at OpenAI”.
Why does OpenAI say they were fired?
For violating policies on handling sensitive information. OpenAI says its investigation found “a significant breach of trust beyond what’s outlined in the letter”, but it has not said publicly what that was.
What does the OpenAI open letter ask for?
Three things: keep the commitment to embed third-party safety auditors, preserve the monitorability of frontier models, and set out clear rules for how staff may work with outside safety organisations.
Has OpenAI agreed to the demands?
Partly. It says it is finalising contracts with outside safety assessors and agrees that monitorability needs an industry-wide commitment. It stands by the firings and has not published new rules on external contact.
References
OpenAI cannot make AI safe on its own (open letter, PDF)
A note from our research leaders (OpenAI Newsroom on X)
Tomek Korbak on his firing (X)
Mikita Balesni shares the letter (X)
Jasmine Wang on her firing (X)
Fired OpenAI safety researchers dispute misconduct claims, warn of chilling effect (TechCrunch)
Trio of workers fired by OpenAI: read their warning letter in full (Newsweek)
Fired OpenAI safety researchers say they were pushed out over ‘suspicious’ circumstances (CNN)
Fired OpenAI safety researchers dispute their dismissals in open letter (Engadget)
OpenAI’s safety firings raise awkward questions (Fortune)
‘Looks like they’re firing whistleblowers’ (Common Dreams)
Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety (arXiv)
OpenAI Hugging Face incident investigation (METR)
AI embedded evaluator jobs (The New Stack)
Employment Rights Act 1996, section 43G (legislation.gov.uk)
Employment Rights Act 1996, section 92 (legislation.gov.uk)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
