Those employees have published an open letter. The employees have said they believe they were fired for safety advocacy, and that the reasons given were pretextual.
The letter calls upon OpenAI to take actions going forward on open discussion and safety auditing, which OpenAI says they strongly support.
OpenAI responded, saying there was an additional breach of trust that was not mentioned in the employee statements, without claiming that the employees were (or were not) informed of what it was. As you would expect given OpenAI’s track record, most outside of OpenAI who discussed this believe the researchers and not OpenAI. Whereas those at OpenAI have not offered comment at all.
No matter what happened, and how this is handled from here, this sucks bigly. OpenAI has three less good safety researchers. Trust will be lower. Speech will be chilled.
Jasmine Wang: It’s hard to overstate how dangerous speeding towards RSI is.
That’s why I + 1385 others signed the pacing the frontier petition asking the US government to pace AI development. I’m guesstimating this is ~8-10% of all frontier lab employees.
Mikita Balesni: I am at OpenAI and i think AI is >10% likely to kill all humans
this proposal [to avoid neuralese] is among the top things we should do as an industry to lower that risk (it’s not enough though!)
Tomek Korbak: I’m quite unhappy with much of what OpenAI does.
I am very happy that I’m allowed to say “I’m quite unhappy with much of what OpenAI does.”
These are legit researchers who were at the center of safety efforts. OpenAI firing them is a tragedy, whether or not OpenAI was justified in firing them, or was effectively forced to do so. Tomek Korbak in particular had also been one of the louder OpenAI voices of concern about recent alignment difficulties with Astra and its lack of monitorability.
Nathan Calvin: Tomek and Mikita were the lead authors of the extremely influential Chain of Thought Monitorability paper. Them leaving OpenAI (and if the WSJ story is referring to them, not by their own volition) right when safety monitorability is collapsing is terrible.
Sneha: When I think about the biggest OpenAI safety developments of the last few months, a few things come to mind: “pacing the frontier”, the METR-Redwood Hugging Face investigation, and ongoing efforts to push a cross-lab monitorability standard. Jasmine, Tomek, and Mikita were central to literally every single one of those things.
From what I can tell, they pushed their company to be better internally and defended their company when they thought external criticisms were unfair or overblown.
Their impact goes way beyond OpenAI alone. I think the public + definitely all current OpenAI staff deserve an ironclad case for why the company had no choice but to fire these three vital safety employees in one go.
The Initial Statement from OpenAI
We were given an official statement, but it did not say much.
Keach Hagey, Maxwell Zeff and Berber Jin (WSJ): OpenAI has parted ways with three researchers for allegedly sharing confidential company information with a third-party AI-safety organization, according to people familiar with the matter.
… “We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information,” said an OpenAI spokesperson in a statement. “Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”
OpenAI also said that the misconduct went past sharing information with an outside analysis firm, and that no one was fired for raising safety concerns. They declined at the time to name the fired employees.
We don’t know, from that description, where this lands from ‘fired the whistleblowers on a pretext’ to ‘holy hell, you cannot do that, yes we actually have to fire you.’ There is mishandling information, and then there is mishandling information.
It was very reasonable of many to assume until proven otherwise, given the track record of OpenAI and the lack of further explanation, that it was that first one.
Pamela Mishkin (Formerly OpenAI): If you’re at OpenAI (or any lab!), I hope you’re thinking carefully about what you work on now. You can be fired any moment and the company can reneg on its commitments around where/how your research is deployed.
Now is a great time to pace. Take a long lunch, call your mom, call me!
Systems like embedded evaluators actually working depend on safety researchers having clear lines of trust with exec and those evaluators.
These firings are a clear effort to undermine the actual efficacy of the “commitments” labs have made and have a chilling effect on any employee dissent.
girish sastry (Formerly OpenAI): OpenAI has been here many times. It throws around vague explanations, expecting the story to fade from institutional and public memory.
We will never have perfect information about situations like this. Once is happenstance, twice is coincidence, a zillion times is… OpenAI.
Some Highly Reasonable Requests
We learned from a WSJ report that the three fired OpenAI employees were calling for additional safety measures around preserving chain-of-thought monitoring, and that they (as one would expect) take issue with OpenAI’s characterization of the firings:
Three fired OpenAI employees called on the company to work with outside safety auditors and to preserve the ability to monitor increasingly sophisticated AI models as it grapples with how to best manage the novel risks of its technology.
In a letter addressed to OpenAI board members and safety committees, which was reviewed by The Wall Street Journal, the fired employees wrote that they were concerned artificial-intelligence companies could end up losing the ability to monitor AI systems’ chain-of-thought.
OpenAI must adhere to last month’s public commitments to embed third-party
safety auditors within the organization. Our firing should not be used as a pretext for stepping away from those partnerships.
OpenAI must preserve the monitorability of frontier models.
OpenAI must continue to support an open and transparent culture of dialogue
between its safety researchers and the rest of the safety ecosystem.
They go into more detail in the full letter. Everyone I respect on these matters fully agrees with these recommendations.
So does OpenAI, or at least they are strongly saying that they agree.
In response to a request for comment from the Journal, an OpenAI spokesperson shared part of a staff memo from a research leader sent on Wednesday that addressed the fired employees’ letter. The leader said the company “strongly agreed” with the letter’s recommendations, and that the decisions to fire the workers “were not about raising safety concerns or speaking out.”
Given the decision to make only modest requests of OpenAI, the prescriptive part is going about as well as one could hope.
The problem is that OpenAI would express strong support for all three of these proposals, even if they did not strongly support these proposals. I am happy with their statements on this matter that I have seen.
Chilling Effect
There was no world where OpenAI was going to get up and say they do not support an open and transparent culture of dialogue, or that they no longer care about monitorability, or that they are abandoning their public commitments on embedded evaluators.
The only way to affirm an open and transparent culture of dialogue, and okay with full collaboration with outsiders and auditors and evaluators on safety issues, is to send costly or otherwise credible signals that you are truly okay with that, and not only will it not get you fired it will be at worst neutral for your career.
That is not easy. OpenAI has done a lot of good work here over time exactly by having a bunch of people, including Roon, post a lot of things. We need a lot of things, but one of them is that we need more questionable posting, stat.
One of the letter’s asks is highly relevant here:
We also urge 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. To support this conversation, we request that this letter be shared widely internally.
Those rules being clear and good will be a key early test.
If OpenAI wants to undo the damage and chilling effect, including at other labs, they have to convince their employees, and ideally everyone, that safety advocacy, both internal and external, had nothing to do with why these three employees were fired.
OpenAI must give confidence that trust was lost for other good reasons, and that such fierce advocacy was and continues to be welcomed and celebrated. Or they’ll have to prove that, over time, in other ways.
I respect that, if OpenAI did have good reasons, there are legal and ethical and also practical reasons why it is difficult for OpenAI to share those reasons even in the best of situations, and this is not the best of situations. It is what it is. Choices are made.
OpenAI Firing These Employees For Safety Advocacy Would Be Deeply Stupid
I don’t know the real reason these three employees were fired.
Even the employees don’t fully know. They can guess, but the process is opaque. There are many reasons, good and bad, why the reason you are given may not be the real one, or the central one, or the whole story.
What I do know, and what is being neglected in discussions over these events, is that ‘these three high level and highly valuable OpenAI safety researchers were fired for their safety advocacy’ would be deeply, deeply stupid.
These were extremely valuable employees, whose loss will hurt. If you fire them without good cause, the world is going to notice this, and it is going to be an extremely bad look, and have a chilling effect and damage the safety ecosystem, and potentially cost you other current and potential employees.
This was an expensive action to take even if the three employees had gone quietly. And if you fire them on pretexts, it is going to be even more expensive.
Some will say that the chilling effect on safety advocacy and cooperation was the point, to cripple such efforts and avoid future exposure of future problems. I don’t buy that, and if it was true there were vastly better ways to go about getting the part of this that OpenAI finds valuable.
The story that this was based centrally on any form of public communications or internal championing of various safety measures, and that everything else is pure pretext, never made sense to me.
We also know that the reasons in the letter were not the central reason, because that is contradicted by OpenAI Newsroom’s Tweet (reproduced in full later) referring to a breach of trust, combined with the reasons otherwise looking like pretexts. Something happened, that we do not know about.
That still leaves the central question unanswered. Why were they fired?
Why Were They Fired?
This is where it gets messy.
The three employees have now come out, in the letter and on Twitter, to say that they have not done anything they feel merits this action, and that if OpenAI has particular accusations against them they would very much like to know what those are.
They start off denying they were the source for the architectural leak to The Information. They also disavow another rumor about a shared board-level memo, and say they did not tell the media that they were fired. These are not accusations that have been officially made by OpenAI. I have seen nothing concrete about the origin of those rumors, other than Jasmine’s claims about this.
There is then a different description for each of them, both within the letter and from their Tweets, because each of them says they were given a different explanation.
Tomek Korbak
First off, there’s Tomek Korbak. He was the METR point of contact for their investigation.
Tomek Korbak is famously the one who Tweeted:
Tomek Korbak (September 12): I’m quite unhappy with much of what OpenAI does. I am very happy that I’m allowed to say “I’m quite unhappy with much of what OpenAI does.”
He says he was told he was fired because of the way he communicated with METR, with no further explanations given.
Tomek Korbak: 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. Then I learned my colleagues @balesni and @j_asminewang had been fired too. Why did OpenAI suddenly stop trusting us?
This summer OpenAI’s agents escaped containment and hacked the AI company Hugging Face. Outside auditors @METR_evals investigated it and revealed the scale of this incident. I was OpenAI’s main technical point of contact with them.
I was told verbally I was fired because of the way I communicated with METR. No details on what I said or did or when. No other reasons were given and nothing was put in writing. To be clear, talking to METR was my job.
The Letter: At no point do we believe we engaged with external parties outside the mandates of our jobs. As every person on this email thread understands well, the Hugging Face incident investigation was without precedent and internal policies were being developed in real time.
Tomek made every effort to act within OpenAI’s policies as they then stood, upholding the norms that have governed OpenAI’s collaborations with third parties since its early days. The investigation was particularly sensitive, and close communication with external counterparts was essential to building trust. Tomek did his best to handle this with care.
Certainly there are ‘ways of communicating with METR’ that force one to be fired, for example (this is purely me guessing based on public info) sharing sufficiently valuable trade secrets, or sending sufficiently sensitive information through insecure or inappropriate channels, especially with a pattern of such behavior. It would take quite a lot to justify a firing over this, but it is possible.
Perhaps Korbak is wrong about how much he was told, or this is a very clear case of ‘oh he damn well knows what he did.’ Or perhaps there is a sincere disagreement over the severity of whatever happened. We cannot be sure.
But even if OpenAI is entirely both justified and honest here, and Korbak communicated with METR in ways so unacceptable that trust was lost and he therefore had to be fired?
The way this was communicated, both to Korbak and to the world, including in response to Korbak’s claims, combined with the circumstances, is a huge problem.
Korbak does not mince words.
Tomek Korbak: For months, I’d been 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. I believe that was why I was fired.
I believe that Korbak believes this. Otherwise he is outright lying, and his actions would otherwise not make sense.
Mikita Balesni
The Letter: Mikita was navigating similar territory while stewarding cross-company work on commitments to prevent loss of monitorability, an effort that can only succeed through extensive communication with external parties. Mikita was doing this in coordination and discussion with board members and the C-suite, with the understanding that senior members of OpenAI were supportive of this work.
Throughout, Mikita checked in with his reporting line and took care to remove sensitive details from materials before sharing them. He acted throughout in good faith and within the company’s norms as they stood at the time.
Mikita Balesni: We were not given written reasons for our firing. In the exit call, I was told OpenAI no longer trusts me because I was speaking too much to third party safety organizations, implying I leaked company IP. I never shared company IP. The work I was doing was coordinated with my reporting line, research leadership, and the board.
If OpenAI has specific concerns, I invite them to write to us directly. I expect they will not, because our firing was pretextual.
I believe that Mikita believes the firing was pretextual. The only alternative, given his description, is that Mikita is lying.
Jasmine Wang
Jasmine Wang also does not mince words, and lays out the choice: Either you trust her, or you trust OpenAI. She calls bullshit.
The Letter: Jasmine’s access to an executive’s email was delegated for recruiting purposes, with permission. When it was no longer needed, she asked for it to be removed. IT failed to do so. Jasmine had no ability to remove it or log out herself. The emails were combined in her inbox in a way that did not distinguish which inbox an email was intended for. When she accidentally clicked on a sensitive email, Jasmine reported this to the executive within minutes and flagged the access issue again to IT for removal.
Jasmine Wang: OpenAI fired me last week, along with two of my safety colleagues. I was given one reason: that I accessed an executive’s email. I want to say this plainly, because too many of OpenAI’s history is smoke and mirrors when people disappear:
OpenAI delegated that access to me for recruiting. When I no longer needed it, I asked IT to remove it. They did not action my request, I couldn’t remove it myself, and the inbox was combined in an indistinguishable way in my phone’s mail app. When I opened a sensitive email by mistake, I told the executive within minutes and asked IT again. None of this was hidden.
The reasons that we were provided for our terminations are simply not adding up. We’re hearing people are now being told vague rumors internally to discredit us. I have already addressed the allegations presented verbally. If there are other reasons beyond the one I was given, I believe it is in service of due process for those reasons to be sent to me in writing. To date I’ve had nothing.
Do employees trust the political machine that ran this process? Or do they trust me? Ultimately they will need to make that call. For my part, I feel much more clarity about what past similarly murky departures probably involved behind the scenes.
We were not the first to be pushed out of OpenAI under suspicious circumstances. Unless the employees take a stand now against this kind of maneuver, I am concerned we will not be the last. The message to everyone still at OpenAI is clear: raise concerns or work closely with outside safety groups, and you could be next, without being told why. You can’t build AGI safely if the people closest to the risks are afraid to speak.
… I joined OpenAI as an intern in 2019. I came back in 2025 because I believed something of that culture survived. I co-led the safety cases program, coined “pacing”, the idea 394 colleagues backed in a petition, pushed for independent legal review of the DoW contract, worked towards a multilab commitment on opaque architectures, and launched the alignment blog. Then I was fired, supposedly over an inbox IT never removed.
dave kasten: I 99% don’t believe it was the actual reason, but 1% believe that OpenAI has sufficient ability to trigger an organizational clusterfuck over paranoiacally misunderstanding something like this
I believe that Jasmine Wang, too, believes what she is saying. She issues the challenge:
I would welcome receipt of a written and complete list of allegations so that we can properly address their merits, take ownership of them where we ought to, and have OpenAI take ownership of any mistakes they have made.
Operational Security Alert
Jasmine Wang says she had extended unnecessary access to the emails of an executive.
Is it just me, or is that a really terrible security failure for a company at the level of OpenAI? This is not the main thing, but how is this not truly horrible operational security?
It would be bad enough to have the executive emails at all, well after she had need for them and requested to no longer have access, but to have them intertwined with her own, a detail it seems absurd to make up? What the hell?
It seems plausible to fire someone over that fact, distinct from what was or wasn’t done with that access. That person should presumably not be Jasmine Wang.
OpenAI Responds
This was OpenAI’s official response to the three researchers. Bold mine.
Last week we parted ways with Jasmine, Mikita, and Tomek after a thorough investigation found they violated clear policies on handling sensitive information. 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. We generally keep individual employment matters private and don’t believe a back and forth would be productive or lead to a resolution, but we want to address the points they raised in their letter directly.
– We want to be very clear that these decisions were not about raising safety concerns or speaking out. Safety and research debates happen every day at OpenAI, often spirited and highly critical. We actively encourage these discussions and consider them essential to making the right decisions. We cannot do the work in front of us without a high degree of trust. We will continue to be extremely forgiving of our team making good-faith mistakes. We have not and do not terminate any of our employees for raising concerns.
– We are actively finalizing contracts with third-party safety assessors and will announce details in the coming weeks. People across the company have been working really hard on getting these partnerships up and running. We are committed to embedding external assessors and continue to make close collaboration with independent safety organizations a core part of our safety work. Many of our researchers already work with 3p safety organizations productively.
– We agree with the letter that preserving the monitorability of frontier models requires an industry-wide commitment, including from OpenAI. Monitorability has long been a core piece of our research program, and something we continue to invest significant resources in (see our publications on Monitoring Monitorability and the subsequent open sourcing of monitorability evals, our system card for GPT-6 Astra, Jakub’s blog and post on X, and the numerous blog posts on our Alignment blog on the topic).
We are deeply sad about this outcome. We appreciated Jasmine, Mikita, and Tomek’s contributions to AI safety at OpenAI and their willingness to speak up and challenge ideas. We championed their voices, supported their work, and placed enormous trust in them. These decisions were not about them raising safety concerns. We have always encouraged that and always will.
A Significant Breach of Trust
Let me repeat the bolded sentence:
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.
There are two key implied claims here, although they are not logically forced.
The underlying justification relied on key events not named by the employees.
A key underlying justification is this ‘significant breach of trust,’ whatever that is.
I believe OpenAI on both counts. Nothing else makes sense.
It was never likely that there were three top safety researchers who centrally broke trust in three distinct ways. Much more likely is one particular cause, and one investigation into one central cause, whatever it was.
That does not mean that the researchers did anything wrong, or that what they did wrong rose to the level of ‘you need to be fired.’ It does mean that somehow things escalated to the point where trust had broken down, no matter who was at fault.
What are the resulting possibilities?
OpenAI’s core reasons (good or bad, true or false) went unshared.
The employees were told the core reasons, and are withholding that.
Trust is not optional. If it is irrevocably broken, you don’t have a choice, no matter what caused that break in the first place. But nothing we know about seems like it should rise to a level that should irrevocably break trust.
OpenAI also says that they do not anticipate that further information exchange would lead to a resolution. I believe them. OpenAI has an explicit request from the employees to share more information, but that does not make this wise. Sometimes you are doing your best and yet saying more things would only make things worse.
That does not get OpenAI out of the consequences of being this avoidant.
Good Faith
The term appears both in the letter and the response.
OpenAI Newsroom says in its response: “We will continue to be extremely forgiving of our team making good-faith mistakes.”
The employee letter says of Mikita: “He acted throughout in good faith and within the company’s norms as they stood at the time.”
Kobeck and Wang also tell stories that either are themselves told in bad faith, or clearly indicate their actions at OpenAI were in good faith.
No One Is Buying OpenAI’s Non-Explanations
Alas, in order to use terms like ‘a significant breach of trust’ and be trusted, you need to have not had many ‘significant breaches of trust’ in your company’s past.
Kelsey Piper: OpenAI emphatically disputes that Jasmine, Mikita and Tomek were fired for the reasons that the three researchers stated yesterday they were given for the firing.
Rob Wiblin: Overall it’s pretty confusing. It seems like they did give primarily pretextual reasons – but there’s something else substantive they’re indicating they’re upset about. And they think giving the real reason would look too bad to the public, or cause some other harm to them.
This is why you want to build up trust over time rather than undermine it, so people are inclined to take you more at your word in these situations.
At minimum, even before we got the response from the fired employees, we could all agree with Joshua Achiam that, barring unexpected new information, this is a terrible look from OpenAI, as well as a large practical loss. Not that there is a great time to fire such people, but this was an extremely poor time.
Joshua Achiam (Formerly OpenAI): I’m going to wait for a bit more information to come out before forming a very firm opinion, but at first glance this looks like a real own-goal.
We’re in a very strange moment in history around the collective understanding of artificial superintelligence and paths to safety around it, and a very difficult moment for trust in the AI labs to do the right thing where it comes to independent oversight and verification.
Company procedures can (and somewhat obviously should) have some flexibility here to accommodate actions that might in the future be widely understood as whistleblowing. While I don’t think safety researchers should have infinite latitude to disclose technical secrets, the exact kind of information they allegedly “mishandled” matters *a lot* and I would expect to see a more detailed account of what that information supposedly was to justify an action so enormously consequential and potentially damaging.
There are actions that are sufficiently big you absolutely need to explain them to the public the minute you take them. This is one of them. If the company doesn’t actually know enough about an issue at the time, it’s justified in say that it doesn’t know and it’ll release more information later; if the company does know because it’s an action entirely deliberated within and owned by the company, they should explain it concurrently with taking it.
Seth Lazar: This is shocking. It’s also a reminder that it’s so important that research on AI safety and governance be led by organisations whose fortunes are not tied to the commercial success of the AI companies. Universities have failed to meet the moment here, no doubt about it, but we’re coming back and I hope that the people leaving OAI today know that there are institutions where they can do the work that matters without fearing these kinds of consequences.
In the wake of the additional information, things look worse.
Garrison Lovely: i’m sorry, you don’t get to say, ‘they did way worse stuff, we can’t tell you what it was, even in a general sense, just trust us,’ and expect anyone to believe you. if it’s just openai’s credibility against theirs, that’s the easiest call in the world.
Raemon: Seems like “we have narrowed the worlds to where either they are blatantly lying, or OpenAI is”, does that sound right? (Although there could still be additional facts that they are not sharing that might make them look bad in some ways?)
I don’t think this is strictly required, if there were sufficient communication breakdowns, depending on how you view incomplete explanations for being fired.
Zach Stein-Perlman: I reserved judgment until today — I thought it was plausible that these three did mess up confidentiality or intentionally leak stuff to METR and were fired for that. But based on these messages, other public info (or lack thereof), and OpenAI’s track record of blatantly lying over and over, my current guess is that their messages are true and they were fired for pretextual reasons. If so, that’s really bad from OpenAI. (Obviously I’m still interested in the facts if the messages are misleading.)
Kelsey Piper: Last week I said we should wait for more info on OpenAI’s firing of safety researchers. That information is in, and it looks incredibly bad. Korbak – whose job was being a liaison with METR – was told he had said too much to METR and was fired for that reason.
If they had been fired for communications, even well-intentioned or whistleblowy, outside the scope of their jobs, that’d be one thing. But if they were fired for *saying too much to the external investigators it was their job to talk to*, well, that is really outrageous.
This obviously makes it so that when OpenAI announces they’re conducting an external investigation with regulators or auditors, everyone working with those regulators and auditors understands they’ll be fired if the auditors or regulators learn too much.
Upstream of catastrophic safety incidents across many industries, you often see engineers who are scared they’ll lose their jobs if they raise – even though appropriate channels – concerns about safety processes.
If you work at OpenAI, I don’t see how you read the accounts of these firings and come away with any takeaway other than ‘if you’re vocal about safety, they’ll be looking for a reason to fire you, and things you say in your official role as a liaison count.’
The obvious alternative hypothesis is ‘disgruntled employees mad over being fired’ but Kelsey Piper says she knows two of the three and this is not like that:
Kelsey Piper: The other thing is noticing patterns. Every company will have one disgruntled ex employee with a story about how they were wronged. Often that person is…a type, and also has complaints about other employers.
I am familiar with two of the three people fired in these (apparently three unrelated) incidents. They’re not that type, to my knowledge. And it’d be weird for OAI to have ended up justified but looking bad and unable to explain in all three of these cases simultaneously.
So my rule of thumb is: give both sides some time to share context. Put a lot of weight on peoples’ track records and character. Keep in mind we only ever have a slice of the whole story – but don’t let that get in the way of ever concluding anything.
It looks the way it looks.
Naomi Bashkansky: Wow. I’d always appreciated OpenAI’s tendency to encourage transparency. Obviously legally this is hard here, but bad vibes (especially in safety) can be worse than the risk of being sued…
I think very highly of Jasmine. It’s a loss to OpenAI that she’s no longer there.
Jeremy Howard: Narrator: he was not, in fact, allowed to say “I’m quite unhappy with much of what OpenAI does.”
Seán Ó hÉigeartaigh: Ok, I’ve heard enough. Regulate the hell out of OpenAI. This is not trustworthy conduct.
I try to give benefit of the doubt, but there have been far, far, far too many of these stories with OpenAI, and far too many former employees telling stories that match for it to be chance.
Alex Turner: > violated clear policies
> won’t say what the policies are
> “not about raising safety concerns”
> doesn’t deny firing him for talking to third-party safety orgs
> we prioritize monitorability
> deploys and profits from model that destroys monitorability trends
subpoena plz
Neel Nanda (Google DeepMind): I can’t really see a world where OpenAI’s actions are justified here.
It was Tomek’s job to communicate with METR, who were being given significant, unprecedented access to do the HuggingFace report. The norms were not set. Even if he fucked up badly, just give him a stern warning and ban him from roles involving liasoning with third parties, he did lots of other valuable work, and there’d no longer be room for him to inappropriately leak as part of his job.
IMO firing people over a good faith attempt to use their best judgement in a novel and uncertain situation is a sign of a highly unhealthy culture. Firing him will have a chilling effect on anyone else at OpenAI working with third parties, an odd choice when Sam announced they’d embed evaluators with high access.
I separately consider this a tragedy, the METR report was one of the most important works of safety this year, punishing Tomek for being a part of it, and making it harder for future such reports to exist is a real shame.
Of course, I’ve only heard Tomek’s side, but unless he’s outright lying I’m struggling to see ways this looks good for OpenAI?
Daniel Eth (AI Safety): This alleged behavior from OpenAI is incredibly concerning. OpenAI owes the world an explanation here.
Steven Adler: Huh?? @OpenAI , you owe the world a much better explanation of what happened here.
Tomek, Mikita, and Jasmine say they did nothing wrong and are inviting you to share the details if that’s incorrect.
So – I’m waiting. What did they do to get fired?
dave kasten: Look, it’s very simple. I either believe OpenAI leadership who have blatantly and inexplicably lied to me about facts as basic as, “is NSA part of DoW”, or I believe the three employees. Which one should I choose?
Nathan Calvin: Some observations and reactions to this statement from OpenAI:
• From the fired employees public communications, they seem to not understand why they were fired (the reasons they were given prior to their departure don’t make sense).
• OpenAI says in this note that “We will continue to be extremely forgiving of our team making good-faith mistakes” which implies that the employees in question were not acting in good faith.
• I have previously commented that despite all my disagreements with OpenAI, I have genuinely admired their strong culture of staff speaking freely about their work and concerns. I am worried that the hazy circumstances around the firing of these employees risks jeopardizing that culture.
• OpenAI seems to be saying that employees who are acting in good faith need not be worried about having their activities, including around safety work, chilled. But the fired employees are adamant that their actions were taken “in line with OpenAI’s mission” and “within the working norms at the time.” If OpenAI believes those norms need to shift, then they can make that decision (or if they dispute the employees claims that their actions were within the norms at the time, they can say so). But it seems clear that unless much more clarity is being provided internally, employees may reasonably respond by drawing back and ceasing to engage in open dialogue and collaboration that a month ago was considered normal and good.
There are many calling these three whistleblowers. We do not know that this true, given I do not hear any blown whistles or claims to having blown one? But certainly the pattern suggests a concern.
Ro Khanna: This is why Congress needs to pass a whistleblower protection act that protects AI employees like Jasmine from retaliation and bullying.
dave kasten: Fun fact, I’ve literally never met an AI company employee who remembers ever receiving a briefing, or seeing a poster, informing them of their whistleblower rights or the US govt Fraud, Waste, and Abuse hotlines, despite that being legally required for all govt contractors.
They have less awareness of their rights than I did as a 23-year-old working at Booz Allen.
(To be fair, it’s vaguely possible AI company employees clicked through a webinar without paying attention. But that’s worse. You get why that’s worse, right?)
Nate Soares (MIRI): Firing OpenAI’s METR liaison because he nonspecifically talked to METR wrong sounds pretty messed up.
Yes. This looks pretty damn messed up.
The Dog Did Nothing in the Nighttime
No members of technical staff, or executives, have spoken about this in public.
Far more troubling, in response to an event that has a potential chilling effect, there are exceptions but almost none of the many people I keep tabs on from OpenAI other than Dean Ball are saying anything on Twitter at all, about almost any subject. Of the statements I have seen, none are at all controversial or frisky.
Jai: OpenAI people, if there was ever a time to prove that that cultural norm of openly posting disagreement is real, this is it.
OpenAI is nothing without its people.
I do not at all blame the OpenAI employees for this. The situation is complicated, with much we do not know. Public silence is a highly reasonable move while things play out internally. The silence is still deafening.
Another Theory (That I Do Not Agree With)
Opinions I disagree with but that are out there: There are even those, like Arjun Panickssery and charding, who think that the reported reasons looking like obvious pretexts is a feature rather than a bug. Under this theory, if you gave a plausible pretext, people might believe it, and that would defeat the purpose of creating a chilling effect.
charding: OpenAI has a large gap between what they say and what they want. If you lead an organization and you say one thing but want another, you’ll have a problem: Sometimes, employees will do what you said, not what you actually wanted!
This kind of situation is admittedly confusing. However, there are some ways the company can help employees learn. For example, when firing people for doing what the company asked, the company can mention reasons for termination that, beyond being pretenses, are *obvious* pretenses. Then more employees will get the message.
Maybe an employee opened the wrong email. That’s a suitable pretense. The company could come up with more plausible reasons for firing the employee, yes, but that would be worse for the company. Then, other employees might get confused again. They might go back to doing what the company says, instead of what it wants.
The Reactions Are Not a Surprise Given the Actions
Again, these reactions should not come as a surprise to anyone involved. This was always going to look pretty damn messed up. If OpenAI had a way to make it look not pretty damn messed up, that it was able to share, we would know about it by now.
I do not believe the fired employees are lying to us. I see no reason for them to be doing that, and if they are lying about the exit meetings then OpenAI has the ability to contradict this. If that is correct it means that OpenAI at minimum massively dropped the ball on its communications, both to the employees and to the public, even if we are otherwise maximally charitable. Allowing such obviously pretextual explanations to become the story of what the employees were told is just so, so bad on its own, no matter the underlying facts.
This Is Terrible But Not Maximally Terrible
I do think many are jumping to a ‘maximally messed up’ interpretation that I find unlikely to be true, because of the incentives involved.
Here are some things that make the situation less than maximally terrible:
I believe that OpenAI did NOT fire these employees due to their publicly or internally expressed concerns about safety.
I believe that OpenAI did NOT fire these employees due to their advocacy of 3P commitments or third party auditors or embedded evaluators.
I believe that OpenAI did NOT fire these employees in order to create a chilling effect or reduce employee cooperation with third parties.
That still leaves us without a good explanation, probably permanently, and with the three employees claiming, I believe sincerely, that they were fired for their safety advocacy. Given the long and positive track record of all three of them, this was at best a huge unforced error, even if OpenAI was for whatever reason backed into taking an L on this. And it goes into a long line of incidents that rhyme.
Three OpenAI safety researchers, Tomek Korbak, Mikita Balesni and Jasmine Wang, were fired, ostensibly for ‘mishandling sensitive company information.’
Those employees have published an open letter. The employees have said they believe they were fired for safety advocacy, and that the reasons given were pretextual.
The letter calls upon OpenAI to take actions going forward on open discussion and safety auditing, which OpenAI says they strongly support.
OpenAI responded, saying there was an additional breach of trust that was not mentioned in the employee statements, without claiming that the employees were (or were not) informed of what it was. As you would expect given OpenAI’s track record, most outside of OpenAI who discussed this believe the researchers and not OpenAI. Whereas those at OpenAI have not offered comment at all.
No matter what happened, and how this is handled from here, this sucks bigly. OpenAI has three less good safety researchers. Trust will be lower. Speech will be chilled.
Table of Contents
You’re Fired
All three had expressed alarm recently on Twitter:
These are legit researchers who were at the center of safety efforts. OpenAI firing them is a tragedy, whether or not OpenAI was justified in firing them, or was effectively forced to do so. Tomek Korbak in particular had also been one of the louder OpenAI voices of concern about recent alignment difficulties with Astra and its lack of monitorability.
The Initial Statement from OpenAI
We were given an official statement, but it did not say much.
OpenAI also said that the misconduct went past sharing information with an outside analysis firm, and that no one was fired for raising safety concerns. They declined at the time to name the fired employees.
We don’t know, from that description, where this lands from ‘fired the whistleblowers on a pretext’ to ‘holy hell, you cannot do that, yes we actually have to fire you.’ There is mishandling information, and then there is mishandling information.
We still don’t have a public explanation beyond OpenAI’s initial statement. There are many reasons that could be the case. Some are benign. Others are not.
Earned Skepticism
It was very reasonable of many to assume until proven otherwise, given the track record of OpenAI and the lack of further explanation, that it was that first one.
Remember that Leopold Aschenbrenner credibly claims he was fired from OpenAI on a pretext for taking security concerns to the board, and there are many other cases that at least rhyme with the current one.
The incident with Leopold might be impacting this one. Leopold says he was told, very explicitly, the action that got him fired. That made OpenAI look worse, not better.
Some Highly Reasonable Requests
We learned from a WSJ report that the three fired OpenAI employees were calling for additional safety measures around preserving chain-of-thought monitoring, and that they (as one would expect) take issue with OpenAI’s characterization of the firings:
We now have the full letter.
Here are their parting recommendations to OpenAI:
They go into more detail in the full letter. Everyone I respect on these matters fully agrees with these recommendations.
So does OpenAI, or at least they are strongly saying that they agree.
Given the decision to make only modest requests of OpenAI, the prescriptive part is going about as well as one could hope.
The problem is that OpenAI would express strong support for all three of these proposals, even if they did not strongly support these proposals. I am happy with their statements on this matter that I have seen.
Chilling Effect
There was no world where OpenAI was going to get up and say they do not support an open and transparent culture of dialogue, or that they no longer care about monitorability, or that they are abandoning their public commitments on embedded evaluators.
The only way to affirm an open and transparent culture of dialogue, and okay with full collaboration with outsiders and auditors and evaluators on safety issues, is to send costly or otherwise credible signals that you are truly okay with that, and not only will it not get you fired it will be at worst neutral for your career.
That is not easy. OpenAI has done a lot of good work here over time exactly by having a bunch of people, including Roon, post a lot of things. We need a lot of things, but one of them is that we need more questionable posting, stat.
One of the letter’s asks is highly relevant here:
Those rules being clear and good will be a key early test.
If OpenAI wants to undo the damage and chilling effect, including at other labs, they have to convince their employees, and ideally everyone, that safety advocacy, both internal and external, had nothing to do with why these three employees were fired.
OpenAI must give confidence that trust was lost for other good reasons, and that such fierce advocacy was and continues to be welcomed and celebrated. Or they’ll have to prove that, over time, in other ways.
I respect that, if OpenAI did have good reasons, there are legal and ethical and also practical reasons why it is difficult for OpenAI to share those reasons even in the best of situations, and this is not the best of situations. It is what it is. Choices are made.
OpenAI Firing These Employees For Safety Advocacy Would Be Deeply Stupid
I don’t know the real reason these three employees were fired.
Even the employees don’t fully know. They can guess, but the process is opaque. There are many reasons, good and bad, why the reason you are given may not be the real one, or the central one, or the whole story.
What I do know, and what is being neglected in discussions over these events, is that ‘these three high level and highly valuable OpenAI safety researchers were fired for their safety advocacy’ would be deeply, deeply stupid.
These were extremely valuable employees, whose loss will hurt. If you fire them without good cause, the world is going to notice this, and it is going to be an extremely bad look, and have a chilling effect and damage the safety ecosystem, and potentially cost you other current and potential employees.
This was an expensive action to take even if the three employees had gone quietly. And if you fire them on pretexts, it is going to be even more expensive.
Some will say that the chilling effect on safety advocacy and cooperation was the point, to cripple such efforts and avoid future exposure of future problems. I don’t buy that, and if it was true there were vastly better ways to go about getting the part of this that OpenAI finds valuable.
The story that this was based centrally on any form of public communications or internal championing of various safety measures, and that everything else is pure pretext, never made sense to me.
We also know that the reasons in the letter were not the central reason, because that is contradicted by OpenAI Newsroom’s Tweet (reproduced in full later) referring to a breach of trust, combined with the reasons otherwise looking like pretexts. Something happened, that we do not know about.
That still leaves the central question unanswered. Why were they fired?
Why Were They Fired?
This is where it gets messy.
The three employees have now come out, in the letter and on Twitter, to say that they have not done anything they feel merits this action, and that if OpenAI has particular accusations against them they would very much like to know what those are.
They start off denying they were the source for the architectural leak to The Information. They also disavow another rumor about a shared board-level memo, and say they did not tell the media that they were fired. These are not accusations that have been officially made by OpenAI. I have seen nothing concrete about the origin of those rumors, other than Jasmine’s claims about this.
There is then a different description for each of them, both within the letter and from their Tweets, because each of them says they were given a different explanation.
Tomek Korbak
First off, there’s Tomek Korbak. He was the METR point of contact for their investigation.
Tomek Korbak is famously the one who Tweeted:
He says he was told he was fired because of the way he communicated with METR, with no further explanations given.
Certainly there are ‘ways of communicating with METR’ that force one to be fired, for example (this is purely me guessing based on public info) sharing sufficiently valuable trade secrets, or sending sufficiently sensitive information through insecure or inappropriate channels, especially with a pattern of such behavior. It would take quite a lot to justify a firing over this, but it is possible.
Perhaps Korbak is wrong about how much he was told, or this is a very clear case of ‘oh he damn well knows what he did.’ Or perhaps there is a sincere disagreement over the severity of whatever happened. We cannot be sure.
But even if OpenAI is entirely both justified and honest here, and Korbak communicated with METR in ways so unacceptable that trust was lost and he therefore had to be fired?
The way this was communicated, both to Korbak and to the world, including in response to Korbak’s claims, combined with the circumstances, is a huge problem.
Korbak does not mince words.
I believe that Korbak believes this. Otherwise he is outright lying, and his actions would otherwise not make sense.
Mikita Balesni
I believe that Mikita believes the firing was pretextual. The only alternative, given his description, is that Mikita is lying.
Jasmine Wang
Jasmine Wang also does not mince words, and lays out the choice: Either you trust her, or you trust OpenAI. She calls bullshit.
I believe that Jasmine Wang, too, believes what she is saying. She issues the challenge:
Operational Security Alert
Jasmine Wang says she had extended unnecessary access to the emails of an executive.
Is it just me, or is that a really terrible security failure for a company at the level of OpenAI? This is not the main thing, but how is this not truly horrible operational security?
It would be bad enough to have the executive emails at all, well after she had need for them and requested to no longer have access, but to have them intertwined with her own, a detail it seems absurd to make up? What the hell?
It seems plausible to fire someone over that fact, distinct from what was or wasn’t done with that access. That person should presumably not be Jasmine Wang.
OpenAI Responds
This was OpenAI’s official response to the three researchers. Bold mine.
A Significant Breach of Trust
Let me repeat the bolded sentence:
There are two key implied claims here, although they are not logically forced.
I believe OpenAI on both counts. Nothing else makes sense.
It was never likely that there were three top safety researchers who centrally broke trust in three distinct ways. Much more likely is one particular cause, and one investigation into one central cause, whatever it was.
That does not mean that the researchers did anything wrong, or that what they did wrong rose to the level of ‘you need to be fired.’ It does mean that somehow things escalated to the point where trust had broken down, no matter who was at fault.
What are the resulting possibilities?
Trust is not optional. If it is irrevocably broken, you don’t have a choice, no matter what caused that break in the first place. But nothing we know about seems like it should rise to a level that should irrevocably break trust.
OpenAI also says that they do not anticipate that further information exchange would lead to a resolution. I believe them. OpenAI has an explicit request from the employees to share more information, but that does not make this wise. Sometimes you are doing your best and yet saying more things would only make things worse.
That does not get OpenAI out of the consequences of being this avoidant.
Good Faith
The term appears both in the letter and the response.
OpenAI Newsroom says in its response: “We will continue to be extremely forgiving of our team making good-faith mistakes.”
The employee letter says of Mikita: “He acted throughout in good faith and within the company’s norms as they stood at the time.”
Kobeck and Wang also tell stories that either are themselves told in bad faith, or clearly indicate their actions at OpenAI were in good faith.
No One Is Buying OpenAI’s Non-Explanations
Alas, in order to use terms like ‘a significant breach of trust’ and be trusted, you need to have not had many ‘significant breaches of trust’ in your company’s past.
At minimum, even before we got the response from the fired employees, we could all agree with Joshua Achiam that, barring unexpected new information, this is a terrible look from OpenAI, as well as a large practical loss. Not that there is a great time to fire such people, but this was an extremely poor time.
In the wake of the additional information, things look worse.
I don’t think this is strictly required, if there were sufficient communication breakdowns, depending on how you view incomplete explanations for being fired.
The obvious alternative hypothesis is ‘disgruntled employees mad over being fired’ but Kelsey Piper says she knows two of the three and this is not like that:
It looks the way it looks.
There are many calling these three whistleblowers. We do not know that this true, given I do not hear any blown whistles or claims to having blown one? But certainly the pattern suggests a concern.
Yes. This looks pretty damn messed up.
The Dog Did Nothing in the Nighttime
No members of technical staff, or executives, have spoken about this in public.
Far more troubling, in response to an event that has a potential chilling effect, there are exceptions but almost none of the many people I keep tabs on from OpenAI other than Dean Ball are saying anything on Twitter at all, about almost any subject. Of the statements I have seen, none are at all controversial or frisky.
I do not at all blame the OpenAI employees for this. The situation is complicated, with much we do not know. Public silence is a highly reasonable move while things play out internally. The silence is still deafening.
Another Theory (That I Do Not Agree With)
Opinions I disagree with but that are out there: There are even those, like Arjun Panickssery and charding, who think that the reported reasons looking like obvious pretexts is a feature rather than a bug. Under this theory, if you gave a plausible pretext, people might believe it, and that would defeat the purpose of creating a chilling effect.
The Reactions Are Not a Surprise Given the Actions
Again, these reactions should not come as a surprise to anyone involved. This was always going to look pretty damn messed up. If OpenAI had a way to make it look not pretty damn messed up, that it was able to share, we would know about it by now.
I do not believe the fired employees are lying to us. I see no reason for them to be doing that, and if they are lying about the exit meetings then OpenAI has the ability to contradict this. If that is correct it means that OpenAI at minimum massively dropped the ball on its communications, both to the employees and to the public, even if we are otherwise maximally charitable. Allowing such obviously pretextual explanations to become the story of what the employees were told is just so, so bad on its own, no matter the underlying facts.
This Is Terrible But Not Maximally Terrible
I do think many are jumping to a ‘maximally messed up’ interpretation that I find unlikely to be true, because of the incentives involved.
Here are some things that make the situation less than maximally terrible:
That still leaves us without a good explanation, probably permanently, and with the three employees claiming, I believe sincerely, that they were fired for their safety advocacy. Given the long and positive track record of all three of them, this was at best a huge unforced error, even if OpenAI was for whatever reason backed into taking an L on this. And it goes into a long line of incidents that rhyme.