1. This had nothing to do with how equity was distributed. This was a clause in an agreement that employees were asked to sign when leaving, not when they joined. As other comments have stated, it's entirely plausible he was unaware. Exit agreements aren't something you expect CEO's to be poring over in detail -- they have HR and lawyers for that.
2. 3. Presumably the company is working on that, and this tweet is the first step? This has become pretty public by now anyways.
4. Have you resigned from every job you've ever had if you made a mistake? If you wrote a design document with a single mistaken requirement? This one clause was one miniscule aspect of running the company out of hundreds of thousands. Demanding resignation is massively disproportionate here.
5. "Reconsider your life" is outside the scope of this.
What was Open AI's legal team afraid of? I can only speculate, and maybe there wasn't a single reason for their LIFETIME clause, but I'm going to guess it has to do with the sources of data used to train chat GPT
This was my immediate reaction as well. The data used to train these systems is inherently really sketchy. And it’s something that any engineer working on it would immediately notice. We certainly did, when building our own models.
There's a few options, OpenAI's list of legal risks is extensive and horrifying.
* For the data sources, we know there's a lot of CSAM in there, and that OpenAI knowingly shipped this data to 3rd party companies to tag it. Time ran an article about their Kenyan parters, who quit on them because "Look at this image. If the image is CSAM, push the button" is a horrifying job.
* There's copyright. AI relies extensively on hiding the true scale of the copyright problem by filtering infringing content out of the AI responses. But the engineers of those systems know the dirty details. (It's dubious that such filtering would bring them back into compliance with fair use; LLMs are paraphrasing machines and copyright filters do not cope well with paraphrasing)
* Discrimination is a shitshow. LLMs discriminate as their dataset isn't reflective of reality, but of what and how we choose to record reality. And as Google's "Diverse Nazis" image generation shows, messing with prompts doesn't work to fix this. It'll be discrimination lawsuits either way. All major AI firms know of this problem, and internally investigate it to avoid gaffes like Google's.
In practice, the boring answer to the question is "all of the above and then some". A big problem for OpenAI is that these are massive problems, for which their engineers could be dragged into a courtroom to testify about.
I don't believe this. Having been CEO of several startups, I know the wording of the equity plan and grant documents are really important. There are plenty of things not worth your time paying attention to, but these documents are too important for that.
I don't think the controversy stems from any provisions in the equity plan and grant documents. The Vox article only talked about the off-boarding NDA, and that's all that Altman is talking about here:
> there was a provision about potential equity cancellation in our previous exit docs; although we never clawed anything back, it should never have been something we had in any documents or communication.
Mind you, Altman could still be lying and that would be very much in character, but this also sounds like a very plausible explanation for the hubbub.
This was in something you were asked to sign when exiting the company, and I can very much imagine that it was something HR or legal thought was a good idea to add in a kind of "better safe than sorry" way, and that the CEO wasn't aware of.
I'm not saying it's necessarily true -- but I do think it falls in the realm of believability.
It wouldn't actually involve any repurchase from what I understand, because these aren't even actual shares (RSU's) to begin with. They're not actual ownership at all. They're some kind of special sharing-of-future-profits agreement.
The setup for this - the clause saying that if you didn't sign an exit agreement within sixty days the company would reclaim your equity - was in the equity plan and grant documents: https://x.com/KelseyTuoc/status/1791584341669396560
Not a lawyer, but "general release of claims" seems like a term of art which I would expect to be a something that would waive your right to sue, and not an agreement that includes a non-disparagement clause.
Equity can refer to stock, options, PPUs, etc. The point is the valley norm is that the vested portion is yours to keep. Some private equity backed companies will be more aggressive and claw back vested, but that’s outside the norm for VC backed companies.
* Is such a non-disparagement clause that claws back your earned compensation legally enforceable, or does it have any legal basis (at least in the state of California)?
* Are non-disclosure agreements enforceable post-employment if you're claiming to be a whistleblower unofficially, i.e. you just go to twitter and rant about your ex-employer?
Once my employer asked us to look over an updated employee handbook. There’s was a bunch of nonsense like “no taking pictures in the office” and “no drinking alcohol in the office”. (The office has kegs in it!)
We called this out and HR was like “it was some template, let me see if I can get it changed” and we went from there.
I dunno if nonsense like that happens at a bigger company like OpenAI but I wouldn’t rule it out, nor would I rule out a CEO not knowing about every detail that HR comes up with.
Are stock options "vested equity"? Afaik they're a derivative, and it's surprising to see this exact phrase both in the OpenAI's response to the vox.com article and in sama's tweet. Also, isn't OpenAI using some sort of profit-sharing units instead of options?
As he says, he should have known, but I think it's plausible that he didn't know every detail of their termination agreements. I would certainly be more comfortable if it was something invented by an overeager lawyer rather than policy directly from the top.
Glad to see an offer to fix it for former employees. Although it's not clear what form that fix would take. If people signed a perpetual non-disparagement clause because they believed their vested equity was under threat, the correct fix would be to void the non-disparagement clause. And even asking for a perpetual non-disparagement agreement in the first place is gross to me, unless it comes with perpetual severance pay. I just re-checked my exit paperwork from Google and Meta and Microsoft and I didn't see anything like that. I don't think that Sam is apologizing for that specifically. He seems to be apologizing about the equity part only. I'd be happy to see clarification here.
Unfortunately I think apologies are not usually the right PR move, because the mob generally doesn't accept them, but as the person who posted the top thread on the previous post, I'm glad that he did apologize despite that.
No, it is not plausible that he didn't. This is the sort of thing that CEOs are very involved in the details of. If an overeager lawyer can add a lifetime non-disparagement clause to the equity plan without the CEO even noticing then you have some very big problems. This isn't some routine HR stuff.
> can add a lifetime non-disparagement clause to the equity plan
I thought we were talking about exit docs, not clauses in the equity program. Do we have any reason to believe that the equity program itself contained language to this effect? The Vox article only talks about the NDA employees sign on the way out, it's only random internet comments that I've seen speculating about the contents of the equity program.
I don’t see how there wouldn’t be a relationship with the exit doc , if part of the NDA is that you lose your equity for insulting the company. He knew about this. There’s no way he only knew about the good bits.
> If an overeager lawyer can add a lifetime non-disparagement clause to the equity plan without the CEO even noticing then you have some very big problems
Yeah, it's one of those "are you an asshole or just incompetent?" PR moments. Altman seems to have chosen "incompetent" but neither is a great look for a CEO.
I can assure you, you are wrong. Sam is very busy running a huge company. It is not surprising, or outside the norm, that he is not reviewing exit documents.
Directly working on this stuff with many many companies of different sizes. I know nothing about OpenAI or Sam in particular — perhaps he is a particular involved CEO and was aware — but I can tell you definitely having a CEO of a company this size paying more attention to the business than this type of stuff is extremely normal.
You are confusing two issues. It is standard practice for companies to require a separation agreement with a release of claims and non disparagement clause in connection with receiving any severance, including keeping a laptop like in item 1 you linked to. Companies are not going to pay you or give you something of value and then have you turn around and sue them or disparage them and cause problems.
Sam’s current tweet is solely about whether vested equity could be clawed back as a penalty for refusing to agree to a non disparagement in connection with a separation if you don’t agree to non disparagement and a release of claims. That is well outside the norms.
Requiring non disparagement is now in violation of labor law per the NLRB as of February 2023. If someone is putting one in front of you, contact the NLRB and an employment attorney to pursue recourse.
Companies can, and do, still seek non-disparagement clauses, just with carve outs that make it clear any protected rights per the NLRA or pursuant to other applicable state laws are carved out.
Companies perform illegal acts all the time, accountability is important. Can’t allow malicious people to get away with malicious behavior when mechanisms (regulatory and legal framework) to keep them in check exist.
We likely see what a just world looks like differently, and that’s fine. If someone has to pay you to be quiet when you leave, they have something material to hide worth pursing.
If you feel like sharing examples of the carve outs you mentioned, please do so I can communicate them to regulators for further triage and actioning.
I was making a descriptive statement, not a normative one. Regardless of your political views, I don’t think you are accurately describing the state of the law.
I also think you are frankly getting caught in black and white thinking. Yes, some companies abuse things like non-disparagement clauses, but they can also be used in a perfectly fine manner.
The carve outs typically say something like “you agree to not disparage the Company, provided nothing herein prohibits you from speaking out about workplace harassment, conduct prohibited by the National Labor Relations Act, or [other speech/laws that local law might say you can’t restrict].”
Terminations are emotionally fraught. People’s egos get hurt and it becomes contentious. A company might want to pay you 3 months severance to end on good terms and avoid a larger dispute, but have it be contingent on a release of claims and non disparagement clause. That does not mean it’s because they “have something to hide”.
For example, these clauses are often mutual. Does the terminated executive have something to hide?
I agree that coercively using NDAs and non-disparagement clauses to hide illegal conduct (harassment, labor law violations) is probably illegal and immoral.
But should OpenAI and Ilya be allowed to agree to a mutual non-disparagement, where both wish each other well, instead of having Sam post publicly about how Ilya is a backstabbing disloyal person that can’t be trusted, and Ilya posting about how Sam is a maniacal power hungry awful boss? These are personality conflicts and workplace disagreements, not illegal conduct.
My comments should not be interpreted as “black and white”, but as a total lack of trust in corporate actors, necessitating robust due diligence and potential government intervention to ensure any constraints imposed are entirely legal and proper. It’s possible the carve outs are legal, and there is no malfeasance. But there could be, and that’s where the work lies, because more often than not (and I believe the evidence around this to be robust), there is malfeasance and corporate actors are not acting in good faith.
My political views are, broadly speaking, “be a decent human, defend and empower other humans”. Understandably, what that means is potentially up for interpretation.
My character assessment of Altman was inaccurate, and my comments speak to a posture needed as a counterparty of his or an org he has a control of.
Also, why is he requiring ex-employees to reach out to him privately to work something out on a case-by-case basis when he could just unilaterally and publicly dissolve that clause for former employees
This seems weird. Of course, CEO is not supposed to know everything, but this particular clause was probably seen as a deal breaker for many candidates, including strategically important top talents who care more about freedom. Unless Sam was completely excluded in the recruiting process, I guess he should've heard complaints against this?
We're talking about a clause in the exit documents, not the onboarding documents. This would never have come up while recruiting, only when people quit or are fired, at which point their complaints probably don't seem important enough to escalate to the CEO.
If this was not in the employment contract, then it's even more problematic. I'm pretty sure this is very likely violation of multiple laws. If some random lawyer added this lifetime clause for their own sake without escalating to the board, then they should be immediately fired. It's possible that Sam was not aware of its illegality and now wanted to minimize his liability though.
I don't know whether Sam Altman personally knew about the clause, but clearly OpenAI as an entity knew - it didn't appear out of thin air, and employees who were leaving were clearly away of it. There was even a bit of a minor hullabaloo on Twitter a week or two ago when another employee left and noted that they gave up their equity because they refused to sign the non-disparagement document.
It’s astonishing how emotional this topic gets on HN.
First, when the original saga happened and now when people leave OpenAI, apparently with a non-disparagement / NDA clause in their severance agreement.
Such an agreement seems very standard[1], and I don’t understand why this is so scandalous.
[1] Edit to add: I have seen several such agreements on both sides, so this is from personal experience
73 comments
[ 3.4 ms ] story [ 140 ms ] threadFrom community darlings to a quickly cratering reputation.
> i did not know this was happening and i should have.
> the team was already in the process of fixing the standard exit paperwork over the past month or so.
The cherry on top? He's cry-typing it like a tumblr apology. Just pathetic.
1. Explain how it happened, since he 100% knew. What CEO does not know how his company's equity is distributed and under what terms?
2. Explain how it will be fixed, in detail.
3. Proactively reach out to all those affected, as per step 2.
4. Resign.
5. Reconsider his life and how he handles feedback, since he has had issue with criticism before: https://news.ycombinator.com/item?id=40396787
2. 3. Presumably the company is working on that, and this tweet is the first step? This has become pretty public by now anyways.
4. Have you resigned from every job you've ever had if you made a mistake? If you wrote a design document with a single mistaken requirement? This one clause was one miniscule aspect of running the company out of hundreds of thousands. Demanding resignation is massively disproportionate here.
5. "Reconsider your life" is outside the scope of this.
I wonder what will actually be discussed when impacted employees talk in private with Sam (or whatever subordinate he delegates this to).
I wouldn’t be surprised if former employees are ambushed during the private meeting.
* For the data sources, we know there's a lot of CSAM in there, and that OpenAI knowingly shipped this data to 3rd party companies to tag it. Time ran an article about their Kenyan parters, who quit on them because "Look at this image. If the image is CSAM, push the button" is a horrifying job.
* There's copyright. AI relies extensively on hiding the true scale of the copyright problem by filtering infringing content out of the AI responses. But the engineers of those systems know the dirty details. (It's dubious that such filtering would bring them back into compliance with fair use; LLMs are paraphrasing machines and copyright filters do not cope well with paraphrasing)
* Discrimination is a shitshow. LLMs discriminate as their dataset isn't reflective of reality, but of what and how we choose to record reality. And as Google's "Diverse Nazis" image generation shows, messing with prompts doesn't work to fix this. It'll be discrimination lawsuits either way. All major AI firms know of this problem, and internally investigate it to avoid gaffes like Google's.
In practice, the boring answer to the question is "all of the above and then some". A big problem for OpenAI is that these are massive problems, for which their engineers could be dragged into a courtroom to testify about.
> there was a provision about potential equity cancellation in our previous exit docs; although we never clawed anything back, it should never have been something we had in any documents or communication.
Mind you, Altman could still be lying and that would be very much in character, but this also sounds like a very plausible explanation for the hubbub.
This was in something you were asked to sign when exiting the company, and I can very much imagine that it was something HR or legal thought was a good idea to add in a kind of "better safe than sorry" way, and that the CEO wasn't aware of.
I'm not saying it's necessarily true -- but I do think it falls in the realm of believability.
A few questions:
* Is such a non-disparagement clause that claws back your earned compensation legally enforceable, or does it have any legal basis (at least in the state of California)?
* Are non-disclosure agreements enforceable post-employment if you're claiming to be a whistleblower unofficially, i.e. you just go to twitter and rant about your ex-employer?
We called this out and HR was like “it was some template, let me see if I can get it changed” and we went from there.
I dunno if nonsense like that happens at a bigger company like OpenAI but I wouldn’t rule it out, nor would I rule out a CEO not knowing about every detail that HR comes up with.
Glad to see an offer to fix it for former employees. Although it's not clear what form that fix would take. If people signed a perpetual non-disparagement clause because they believed their vested equity was under threat, the correct fix would be to void the non-disparagement clause. And even asking for a perpetual non-disparagement agreement in the first place is gross to me, unless it comes with perpetual severance pay. I just re-checked my exit paperwork from Google and Meta and Microsoft and I didn't see anything like that. I don't think that Sam is apologizing for that specifically. He seems to be apologizing about the equity part only. I'd be happy to see clarification here.
Unfortunately I think apologies are not usually the right PR move, because the mob generally doesn't accept them, but as the person who posted the top thread on the previous post, I'm glad that he did apologize despite that.
I thought we were talking about exit docs, not clauses in the equity program. Do we have any reason to believe that the equity program itself contained language to this effect? The Vox article only talks about the NDA employees sign on the way out, it's only random internet comments that I've seen speculating about the contents of the equity program.
Yeah, it's one of those "are you an asshole or just incompetent?" PR moments. Altman seems to have chosen "incompetent" but neither is a great look for a CEO.
1) Altman's companies have had similar clauses before: https://news.ycombinator.com/item?id=40396787
2) The entire OpenAI board debacle started because Sam wanted Helen Toner removed from the board for publishing a paper he felt was disparaging to the company: https://thezvi.substack.com/p/openai-the-battle-of-the-board
Sam’s current tweet is solely about whether vested equity could be clawed back as a penalty for refusing to agree to a non disparagement in connection with a separation if you don’t agree to non disparagement and a release of claims. That is well outside the norms.
https://www.nlrb.gov/news-outreach/news-story/board-rules-th... (“Board Rules that Employers May Not Offer Severance Agreements Requiring Employees to Broadly Waive Labor Law Rights”)
https://www.axios.com/2023/03/27/labor-board-says-non-dispar... (“Labor Board says non-disparagement clauses are unlawful — here’s what that means”)
We likely see what a just world looks like differently, and that’s fine. If someone has to pay you to be quiet when you leave, they have something material to hide worth pursing.
If you feel like sharing examples of the carve outs you mentioned, please do so I can communicate them to regulators for further triage and actioning.
I also think you are frankly getting caught in black and white thinking. Yes, some companies abuse things like non-disparagement clauses, but they can also be used in a perfectly fine manner.
The carve outs typically say something like “you agree to not disparage the Company, provided nothing herein prohibits you from speaking out about workplace harassment, conduct prohibited by the National Labor Relations Act, or [other speech/laws that local law might say you can’t restrict].”
Terminations are emotionally fraught. People’s egos get hurt and it becomes contentious. A company might want to pay you 3 months severance to end on good terms and avoid a larger dispute, but have it be contingent on a release of claims and non disparagement clause. That does not mean it’s because they “have something to hide”.
For example, these clauses are often mutual. Does the terminated executive have something to hide?
I agree that coercively using NDAs and non-disparagement clauses to hide illegal conduct (harassment, labor law violations) is probably illegal and immoral.
But should OpenAI and Ilya be allowed to agree to a mutual non-disparagement, where both wish each other well, instead of having Sam post publicly about how Ilya is a backstabbing disloyal person that can’t be trusted, and Ilya posting about how Sam is a maniacal power hungry awful boss? These are personality conflicts and workplace disagreements, not illegal conduct.
My political views are, broadly speaking, “be a decent human, defend and empower other humans”. Understandably, what that means is potentially up for interpretation.
My character assessment of Altman was inaccurate, and my comments speak to a posture needed as a counterparty of his or an org he has a control of.
So it could well have just been part of the template that their legal team likely started with.
That is an extraordinary clause, which is evident by the wide, stunned reaction and the CEOs rapid backpedaling as it becomes public knowledge.
First, when the original saga happened and now when people leave OpenAI, apparently with a non-disparagement / NDA clause in their severance agreement.
Such an agreement seems very standard[1], and I don’t understand why this is so scandalous.
[1] Edit to add: I have seen several such agreements on both sides, so this is from personal experience