I am confident, on the basis of private information I can’t share, that Anthropic has asked at least some employees to sign similar non-disparagement agreements that are covered by non-disclosure agreements as OpenAI did.
Or to put things into more plain terms:
I am confident that Anthropic has offered at least one employee significant financial incentive to promise to never say anything bad about Anthropic, or anything that might negatively affect its business, and to never tell anyone about their commitment to do so.
I am not aware of Anthropic doing anything like withholding vested equity the way OpenAI did, though I think the effect on discourse is similarly bad.
I of course think this is quite sad and a bad thing for a leading AI capability company to do, especially one that bills itself on being held accountable by its employees and that claims to prioritize safety in its plans.
Hey all, Anthropic cofounder here. I wanted to clarify Anthropic’s position on non-disparagement agreements:
We have never tied non-disparagement agreements to vested equity: this would be highly unusual. Employees or former employees never risked losing their vested equity for criticizing the company.
We historically included standard non-disparagement terms by default in severance agreements, and in some non-US employment contracts. We’ve since recognized that this routine use of non-disparagement agreements, even in these narrow cases, conflicts with our mission. Since June 1st we’ve been going through our standard agreements and removing these terms.
Anyone who has signed a non-disparagement agreement with Anthropic is free to state that fact (and we regret that some previous agreements were unclear on this point). If someone signed a non-disparagement agreement in the past and wants to raise concerns about safety at Anthropic, we welcome that feedback and will not enforce the non-disparagement agreement.
In other words— we’re not here to play games with AI safety using legal contracts. Anthropic’s whole reason for existing is to increase the chance that AI goes well, and spur a race to the top on AI safety.
Some other examples of things we’ve needed to adjust from the standard corporate boilerplate to ensure compatibility with our mission: (1) replacing standard shareholder governance with the Long Term Benefit Trust and (2) supplementing standard risk management with the Responsible Scaling Policy. And internally, we have an anonymous RSP non-compliance reporting line so that any employee can raise concerns about issues like this without any fear of retaliation.
Please keep up the pressure on us and other AI developers: standard corporate best practices won’t cut it when the stakes are this high. Our goal is to set a new standard for governance in AI development. This includes fostering open dialogue, prioritizing long-term safety, making our safety practices transparent, and continuously refining our practices to align with our mission.
You should publicly confirm that your old policy don’t meaningfully advance the frontier with a public launch has been replaced by your RSP, if that’s true, and otherwise clarify your policy.
You take credit for the LTBT (e.g. here) but you haven’tpublished enough to show that it’s effective. You should publish the Trust Agreement, clarify these ambiguities, and make accountability-y commitments like if major changes happen to the LTBT we’ll quickly tell the public.
(Reminder that a year ago you committed to establish a bug bounty program (for model issues) or similar but haven’t. But I don’t think bug bounties are super important.)
[Edit: bug bounties are also mentioned in your RSP—in association with ASL-2—but not explicitly committed to.]
(Sidenote: it seems Sam was kind of explicitly asking to be pressured, so your comment seems legit :) But I also think that, had Sam not done so, I would still really appreciate him showing up and responding to Oli’s top-level post, and I think it should be fine for folks from companies to show up and engage with the topic at hand (NDAs), without also having to do a general AMA about all kinds of other aspects of their strategy and policies. If Zach’s questions do get very upvoted, though, it might suggest there’s demand for some kind of Anthropic AMA event.)
Anyone who has signed a non-disparagement agreement with Anthropic is free to state that fact (and we regret that some previous agreements were unclear on this point) [emphasis added]
This seems as far as I can tell a straightforward lie?
I am very confident that the non-disparagement agreements you asked at least one employee to sign were not ambiguous, and very clearly said that the non-disparagement clauses could not be mentioned.
To reiterate what I know to be true: Employees of Anthropic were asked to sign non-disparagement agreements with a commitment to never tell anyone about the presence of those non-disparagement agreements. There was no ambiguity in the agreements that I have seen.
@Sam McCandlish: Please clarify what you meant to communicate by the above, which I interpreted as claiming that there was merely ambiguity in previous agreements about whether the non-disparagement agreements could be disclosed, which seems to me demonstrably false.
I can confirm that my concealed non-disparagement was very explicit that I could not discuss the existence or terms of the agreement, I don’t see any way I could be misinterpreting this. (but I have now kindly been released from it!)
EDIT: It wouldn’t massively surprise me if Sam just wasn’t aware of its existence though
We’re not claiming that Anthropic never offered a confidential non-disparagement agreement. What we are saying is: everyone is now free to talk about having signed a non-disparagement agreement with us, regardless of whether there was a non-disclosure previously preventing it. (We will of course continue to honor all of Anthropic’s non-disparagement and non-disclosure obligations, e.g. from mutual agreements.)
If you’ve signed one of these agreements and have concerns about it, please email hr@anthropic.com.
Hmm, I feel like you didn’t answer my question. Can you confirm that Anthropic has asked at least some employees to sign confidential non-disparagement agreements?
I think your previous comment pretty strongly implied that you think you did not do so (i.e. saying any previous agreements were merely “unclear” I think pretty clearly implies that none of them did include a non-ambiguous confidential non-disparagement agreement). I want to it to be confirmed and on the record that you did, so I am asking you to say so clearly.
I really think the above was meant to imply that the non disparagement agreements were merely unclear on whether they were covered by a non disclosure clause (and I would be happy to take bets on how a randomly selected reader would interpret it).
My best guess is Sam was genuinely confused on this and that there are non disparagement agreements with Anthropic that clearly are not covered by such clauses.
EDIT: Anthropic have kindly released me personally from my entire concealed non-disparagement, not just made a specific safety exception. Their position on other employees remains unclear, but I take this as a good sign
If someone signed a non-disparagement agreement in the past and wants to raise concerns about safety at Anthropic, we welcome that feedback and will not enforce the non-disparagement agreement.
Thanks for this update! To clarify, are you saying that you WILL enforce existing non disparagements for everything apart from safety, but you are specifically making an exception for safety?
this routine use of non-disparagement agreements, even in these narrow cases, conflicts with our mission
Given this part, I find this surprising. Surely if you think it’s bad to ask future employees to sign non disparagements you should also want to free past employees from them too?
This comment appears to respond to habryka, but doesn’t actually address what I took to be his two main points—that Anthropic was using NDAs to cover non-disparagement agreements, and that they were applying significant financial incentive to pressure employees into signing them.
We historically included standard non-disparagement agreements by default in severance agreements
Were these agreements subject to NDA? And were all departing employees asked to sign them, or just some? If the latter, what determined who was asked to sign?
Anyone who has signed a non-disparagement agreement with Anthropic is free to state that fact (and we regret that some previous agreements were unclear on this point).
I’m curious as to why it took you (and therefore Anthropic) so long to make it common knowledge (or even public knowledge) that Anthropic used non-disparagement contracts as a standard and was also planning to change its standard agreements.
The right time to reveal this was when the OpenAI non-disparagement news broke, not after Habryka connects the dots and builds social momentum for scrutiny of Anthropic.
that Anthropic used non-disparagement contracts as a standard and was also planning to change its standard agreements.
I do want to be clear that a major issue is that Anthropic used non-disparagement agreements that were covered by non-disclosure agreements. I think that’s an additionally much more insidious thing to do, that contributed substantially to the harm caused by the OpenAI agreements, and I think is important fact to include here (and also makes the two situations even more analogous).
Thank you for responding! (I have more comments and questions but figured I would shoot off one quick question which is easy to ask)
We’ve since recognized that this routine use of non-disparagement agreements, even in these narrow cases, conflicts with our mission
Can you clarify what you mean by “even in these narrow cases”? If I am understanding you correctly, you are saying that you were including a non-disparagement clause by default in all of your severance agreements, which sounds like the opposite of narrow (edit: though as Robert points out it depends on what fraction of employees get offered any kind of severance, which might be most, or might be very few).
I agree that it would have technically been possible for you to also include such an agreement on start of employment, but that would have been very weird, and not even OpenAI did that.
I think using the sentence “even in these narrow cases” seems inappropriate given that (if I am understanding you correctly) all past employees were affected by these agreements. I think it would be good to clarify what fraction of past employees were actually offered these agreements.
Severance agreements typically aren’t offered to all departing employees, but usually only those that are fired or laid off. We know that not all past employees were affected by these agreements, because Ivan claims to not have been offered such an agreement, and he left[1] in mid-2023, which was well before June 1st.
Ah, fair, that would definitely make the statement substantially more accurate.
@Sam McCandlish: Could you clarify whether severance agreements were also offered to voluntarily departing employees, and if so, under which conditions?
To expand on my “that’s a crux”: if the non-disparagement+NDA clauses are very standard, such that they were included in a first draft by an attorney without prompting and no employee ever pushed back, then I would think this was somewhat less bad.
It would still be somewhat bad, because Anthropic should be proactive about not making those kinds of mistakes. I am confused about what level of perfection to demand from Anthropic, considering the stakes.
And if non-disparagement is often used, but Anthropic leadership either specified its presence or its form, that would seem quite bad to me, because mistakes of commision here are more evidence of poor decisionmaking than mistakes of omission. If Anthropic leadership decided to keep the clause when a departing employee wanted to remove the clause, that would similarly seem quite bad to me.
I think that both these clauses are very standard in such agreements. Both severance letter templates I was given for my startup, one from a top-tier SV investor’s HR function and another from a top-tier SV law firm, had both clauses. When I asked Claude, it estimated 70-80% of startups would have a similar non-disparagement clause and 80-90% would have a similar confidentiality-of-this-agreement’s-terms clause. The three top Google hits for “severance agreement template” all included those clauses.
These generally aren’t malicious. Terminations get messy and departing employees often have a warped or incomplete picture of why they were terminated–it’s not a good idea to tell them all those details, because that adds liability, and some of those details are themselves confidential about other employees. Companies view the limitation of liability from release of various wrongful termination claims as part of the value they’re “purchasing” by offering severance–not because those claims would succeed, but because it’s expensive to explain in court why they’re justified. But the expenses disgruntled ex-employees can cause is not just legal, it’s also reputational. You usually don’t know which ex-employee will get salty and start telling their side of the story publicly, where you can’t easily respond with your side without opening up liability. Non-disparagement helps cover that side of it. And if you want to disparage the company, in a standard severance letter that doesn’t claw back vested equity, hey, you’re free to just not sign it–it’s likely only a bonus few weeks/months’ salary that you didn’t yet earn on the line, not the value of all the equity you had already vested. We shouldn’t conflate the OpenAI situation with Anthropic’s given the huge difference in stakes.
Confidentiality clauses are standard because they prevent other employees from learning the severance terms and potentially demanding similar treatment in potentially dissimilar situations, thus helping the company control costs and negotiations in future separations. They typically cover the entire agreement and are mostly about the financial severance terms. I imagine that departing employees who cared could’ve ask the company for a carve-out on the confidentiality for the non-disparagement clause as a very minor point of negotiation.
It’s great that Anthropic is taking steps to make these docs more departing-employee-friendly. I wouldn’t read too much into that the docs were like this in the first place (as this wasn’t on cultural radars until very recently) or that they weren’t immediately changed (legal stuff takes time and this was much smaller in scope than in the OpenAI case).
Example clauses in default severance letter from my law firm:
7. Non-Disparagement. You agree that you will not make any false, disparaging or derogatory statements to any media outlet, industry group, financial institution or current or former employees, consultants, clients or customers of the Company, regarding the Company, including with respect to the Company, its directors, officers, employees, agents or representatives or about the Company’s business affairs and financial condition.
11. Confidentiality. To the extent permitted by law, you understand and agree that as a condition for payment to you of the severance benefits herein described, the terms and contents of this letter agreement, and the contents of the negotiations and discussions resulting in this letter agreement, shall be maintained as confidential by you and your agents and representatives and shall not be disclosed except to the extent required by federal or state law or as otherwise agreed to in writing by the Company.
And internally, we have an anonymous RSP non-compliance reporting line so that any employee can raise concerns about issues like this without any fear of retaliation.
Are you able to elaborate on how this works? Are there any other details about this publicly, couldn’t find more detail via a quick search.
Some specific qs I’m curious about: (a) who handles the anonymous complaints, (b) what is the scope of behavior explicitly (and implicitly re: cultural norms) covered here, (c) handling situations where a report would deanonymize the reporter (or limit them to a small number of people)?
OK, let’s imagine I had a concern about RSP noncompliance, and felt that I needed to use this mechanism.
(in reality I’d just post in whichever slack channel seemed most appropriate; this happens occasionally for “just wanted to check...” style concerns and I’m very confident we’d welcome graver reports too. Usually that’d be a public channel; for some compartmentalized stuff it might be a private channel and I’d DM the team lead if I didn’t have access. I think we have good norms and culture around explicitly raising safety concerns and taking them seriously.)
As I understand it, I’d:
Remember that we have such a mechanism and bet that there’s a shortcut link. Fail to remember the shortlink name (reports? violations?) and search the list of “rsp-” links; ah, it’s rsp-noncompliance. (just did this, and added a few aliases)
That lands me on the policy PDF, which explains in two pages the intended scope of the policy, who’s covered, the proceedure, etc. and contains a link to the third-party anonymous reporting platform. That link is publicly accessible, so I could e.g. make a report from a non-work device or even after leaving the company.
I write a report on that platform describing my concerns[1], optionally uploading documents etc. and get a random password so I can log in later to give updates, send and receive messages, etc.
The report by default goes to our Responsible Scaling Officer, currently Sam McCandlish. If I’m concerned about the RSO or don’t trust them to handle it, I can instead escalate to the Board of Directors (current DRI Daniella Amodei)
Investigation and resolution obviously depends on the details of the noncompliance concern.
There are other (pretty standard) escalation pathways for concerns about things that aren’t RSP noncompliance. There’s not much we can do about the “only one person could have made this report” problem beyond the included strong commitments to non-retaliation, but if anyone has suggestions I’d love to hear them.
Good that it’s clear who it goes to, though if I was an anthropic I’d want an option to escalate to a board member who isn’t Dario or Daniella, in case I had concerns related to the CEO
Makes sense—if I felt I had to use an anonymous mechanism, I can see how contacting Daniela about Dario might be uncomfortable. (Although to be clear I actually think that’d be fine, and I’d also have to think that Sam McCandlish as responsible scaling officer wouldn’t handle it)
If I was doing this today I guess I’d email another board member; and I’ll suggest that we add that as an escalation option.
Are there currently board members who are meaningfully separated in terms of incentive-alignment with Daniella or Dario? (I don’t know that it’s possible for you to answer in a way that’d really resolve my concerns, given what sort of information is possible to share. But, “is there an actual way to criticize Dario and/or Daniella in a way that will realistically be given a fair hearing by someone who, if appropriate, could take some kind of action” is a crux of mine)
Absent evidence to the contrary, for any organization one should assume board members were basically selected by the CEO. So hard to get assurance about true independence, but it seems good to at least to talk to someone who isn’t a family member/close friend.
(Jay Kreps was formally selected by the LTBT. I think Yasmin Razavi was selected by the Series C investors. It’s not clear how involved the leadership/Amodeis were in those selections. The three remaining members of the LTBT appear independent, at least on cursory inspection.)
I think that personal incentives is an unhelpful way to try and think about or predict board behavior (for Anthropic and in general), but you can find the current members of our board listed here.
Is there an actual way to criticize Dario and/or Daniela in a way that will realistically be given a fair hearing by someone who, if appropriate, could take some kind of action?
For whom to criticize him/her/them about what? What kind of action are you imagining? For anything I can imagine actually coming up, I’d be personally comfortable raising it directly with either or both of them in person or in writing, and believe they’d give it a fair hearing as well as appropriate follow-up. There are also standard company mechanisms that many people might be more comfortable using (talk to your manager or someone responsible for that area; ask a maybe-anonymous question in various fora; etc). Ultimately executives are accountable to the board, which will be majority appointed by the long-term benefit trust from late this year.
As a point of clarification: is it correct that the first quoted statement above should be read as “at least one employee” in line with the second quoted statement? (When I first read it, I parsed it as “all employees” which was very confusing since I carefully read my contract both before signing and a few days ago (before posting this comment) and I’m pretty sure there wasn’t anything like this in there.)
(I’m a full-time employee at Anthropic.)
I carefully read my contract both before signing and a few days ago [...] there wasn’t anything like this in there.
Current employees of OpenAI also wouldn’t yet have signed or even known about the non-disparagement agreement that is part of “general release” paperwork on leaving the company. So this is only evidence about some ways this could work at Anthropic, not others.
I am disappointed. Using nondisparagement agreements seems bad to me, especially if they’re covered by non-disclosure agreements, especially if you don’t announce that you might use this.
My ask-for-Anthropic now is to explain the contexts in which they have asked or might ask people to incur nondisparagement obligations, and if those are bad, release people and change policy accordingly. And even if nondisparagement obligations can be reasonable, I fail to imagine how non-disclosure obligations covering them could be reasonable, so I think Anthropic should at least do away with the no-disclosure-of-nondisparagement obligations.
Neither is Daniel Kokotajlo. Context and wording strongly suggest that what you mean is that you weren’t ever offered paperwork with such an agreement and incentives to sign it, but there remains a slight ambiguity on this crucial detail.
I agree that this kind of legal contract is bad, and Anthropic should do better. I think there are a number of aggrevating factors which made the OpenAI situation extrodinarily bad, and I’m not sure how much these might obtain regarding Anthropic (at least one comment from another departing employee about not being offered this kind of contract suggest the practice is less widespread).
-amount of money at stake -taking money, equity or other things the employee believed they already owned if the employee doesn’t sign the contract, vs. offering them something new (IANAL but in some cases, this could be a felony “grand theft wages” under California law if a threat to withhold wages for not signing a contract is actually carried out, what kinds of equity count as wages would be a complex legal question) -is this offered to everyone, or only under circumstances where there’s a reasonable justification? -is this only offered when someone is fired or also when someone resigns? -to what degree are the policies of offering contracts concealed from employees? -if someone asks to obtain legal advice and/or negotiate before signing, does the company allow this? -if this becomes public, does the company try to deflect/minimize/only address issues that are made publically, or do they fix the whole situation? -is this close to “standard practice” (which doesn’t make it right, but makes it at least seem less deliberately malicious), or is it worse than standard practice? -are there carveouts that reduce the scope of the non-disparagement clause (explicitly allow some kinds of speech, overriding the non-disparagement)? -are there substantive concerns that the employee has at the time of signing the contract, that the agreement would prevent discussing? -are there other ways the company could retaliate against an employee/departing employee who challenges the legality of contract?
I think with termination agreements on being fired there’s often 1. some amount of severance offered 2. a clause that says “the terms and monetary amounts of this agreement are confidential” or similar. I don’t know how often this also includes non-disparagement. I expect that most non-disparagement agreements don’t have a term or limits on what is covered.
I think a steelman of this kind of contract is: Suppose you fire someone, believe you have good reasons to fire them, and you think that them loudly talking about how it was unfair that you fired them would unfairly harm your company’s reputation. Then it seems somewhat reasonable to offer someone money in exchange for “don’t complain about being fired”. The person who was fired can then decide whether talking about it is worth more than the money being offered.
However, you could accomplish this with a much more limited contract, ideally one that lets you disclose “I signed a legal agreement in exchange for money to not complain about being fired”, and doesn’t cover cases where “years later, you decide the company is doing the wrong thing based on public information and want to talk about that publically” or similar.
I think it is not in the nature of most corporate lawyers to think about “is this agreement giving me too much power?” and most employees facing such an agreement just sign it without considering negotiating or challenging the terms.
For any future employer, I will ask about their policies for termination contracts before I join (as this is when you have the most leverage, if they give you an offer they want to convince you to join).
This is true. I signed a concealed non-disparagement when I left Anthropic in mid 2022. I don’t have clear evidence this happened to anyone else (but that’s not strong evidence of absence). More details here
EDIT: I should also clarify that I personally don’t think Anthropic acted that badly, and recommend reading about what actually happened before forming judgements. I do not think I am the person referred to in Habryka’s comment.
I am including both in this reference class (i.e. when I say employee above, it refers to both present employees and employees who left at some point). I am intentionally being broad here to preserve more anonymity of my sources.
Not sure how to interpret the “agree” votes on this comment. If someone is able to share that they agree with the core claim because of object-level evidence, I am interested. (Rather than agreeing with the claim that this state of affairs is “quite sad”.)
A LOT depends on the details of WHEN the employees make the agreement, and the specifics of duration and remedy, and the (much harder to know) the apparent willingness to enforce on edge cases.
“significant financial incentive to promise” is hugely different from “significant financial loss for choosing not to promise”. MANY companies have such things in their contracts, and they’re a condition of employment. And they’re pretty rarely enforced. That’s a pretty significant incentive, but it’s prior to investment, so it’s nowhere near as bad.
(Not answering this question since I think it would leak too many bits on confidential stuff. In general I will be a bit hesitant to answer detailed questions on this, or I might take a long while to think about what to say before I answer, which I recognize is annoying, but I think is the right tradeoff in this situation)
I’m kind of concerned about the ethics of someone signing a contract and then breaking it to anonymously report what’s going on (if that’s what your private source did). I think there’s value from people being able to trust each others’ promises about keeping secrets, and as much as I’m opposed to Anthropic’s activities, I’d nevertheless like to preserve a norm of not breaking promises.
Can you confirm or deny whether your private information comes from someone who was under a contract not to give you that private information? (I completely understand if the answer is no.)
I think this is a reasonable question to ask. I will note that in this case, if your guess is right about what happened, the breaking of the agreement is something that it turned out the counterparty endorsed, or at least, after the counterparty became aware of the agreement, they immediately lifted it.
I still think there’s something to maintaining all agreements regardless of context, but I do genuinely think it matters here if you (accurately) expect the entity you’ve made the secret agreement with would likely retract it if they found out about it.
(Disclaimer that I have no private info about this specific situation.)
I am confident, on the basis of private information I can’t share, that Anthropic has asked at least some employees to sign similar non-disparagement agreements that are covered by non-disclosure agreements as OpenAI did.
Or to put things into more plain terms:
I am confident that Anthropic has offered at least one employee significant financial incentive to promise to never say anything bad about Anthropic, or anything that might negatively affect its business, and to never tell anyone about their commitment to do so.
I am not aware of Anthropic doing anything like withholding vested equity the way OpenAI did, though I think the effect on discourse is similarly bad.
I of course think this is quite sad and a bad thing for a leading AI capability company to do, especially one that bills itself on being held accountable by its employees and that claims to prioritize safety in its plans.
Hey all, Anthropic cofounder here. I wanted to clarify Anthropic’s position on non-disparagement agreements:
We have never tied non-disparagement agreements to vested equity: this would be highly unusual. Employees or former employees never risked losing their vested equity for criticizing the company.
We historically included standard non-disparagement terms by default in severance agreements, and in some non-US employment contracts. We’ve since recognized that this routine use of non-disparagement agreements, even in these narrow cases, conflicts with our mission. Since June 1st we’ve been going through our standard agreements and removing these terms.
Anyone who has signed a non-disparagement agreement with Anthropic is free to state that fact (and we regret that some previous agreements were unclear on this point). If someone signed a non-disparagement agreement in the past and wants to raise concerns about safety at Anthropic, we welcome that feedback and will not enforce the non-disparagement agreement.
In other words— we’re not here to play games with AI safety using legal contracts. Anthropic’s whole reason for existing is to increase the chance that AI goes well, and spur a race to the top on AI safety.
Some other examples of things we’ve needed to adjust from the standard corporate boilerplate to ensure compatibility with our mission: (1) replacing standard shareholder governance with the Long Term Benefit Trust and (2) supplementing standard risk management with the Responsible Scaling Policy. And internally, we have an anonymous RSP non-compliance reporting line so that any employee can raise concerns about issues like this without any fear of retaliation.
Please keep up the pressure on us and other AI developers: standard corporate best practices won’t cut it when the stakes are this high. Our goal is to set a new standard for governance in AI development. This includes fostering open dialogue, prioritizing long-term safety, making our safety practices transparent, and continuously refining our practices to align with our mission.
OK:
You should publicly confirm that your old policy don’t meaningfully advance the frontier with a public launch has been replaced by your RSP, if that’s true, and otherwise clarify your policy.
You take credit for the LTBT (e.g. here) but you haven’t published enough to show that it’s effective. You should publish the Trust Agreement, clarify these ambiguities, and make accountability-y commitments like if major changes happen to the LTBT we’ll quickly tell the public.
(Reminder that a year ago you committed to establish a bug bounty program (for model issues) or similar but haven’t. But I don’t think bug bounties are super important.)
[Edit: bug bounties are also mentioned in your RSP—in association with ASL-2—but not explicitly committed to.]
(Good job in many areas.)
(Sidenote: it seems Sam was kind of explicitly asking to be pressured, so your comment seems legit :)
But I also think that, had Sam not done so, I would still really appreciate him showing up and responding to Oli’s top-level post, and I think it should be fine for folks from companies to show up and engage with the topic at hand (NDAs), without also having to do a general AMA about all kinds of other aspects of their strategy and policies. If Zach’s questions do get very upvoted, though, it might suggest there’s demand for some kind of Anthropic AMA event.)
This seems as far as I can tell a straightforward lie?
I am very confident that the non-disparagement agreements you asked at least one employee to sign were not ambiguous, and very clearly said that the non-disparagement clauses could not be mentioned.
To reiterate what I know to be true: Employees of Anthropic were asked to sign non-disparagement agreements with a commitment to never tell anyone about the presence of those non-disparagement agreements. There was no ambiguity in the agreements that I have seen.
@Sam McCandlish: Please clarify what you meant to communicate by the above, which I interpreted as claiming that there was merely ambiguity in previous agreements about whether the non-disparagement agreements could be disclosed, which seems to me demonstrably false.
I can confirm that my concealed non-disparagement was very explicit that I could not discuss the existence or terms of the agreement, I don’t see any way I could be misinterpreting this. (but I have now kindly been released from it!)
EDIT: It wouldn’t massively surprise me if Sam just wasn’t aware of its existence though
We’re not claiming that Anthropic never offered a confidential non-disparagement agreement. What we are saying is: everyone is now free to talk about having signed a non-disparagement agreement with us, regardless of whether there was a non-disclosure previously preventing it. (We will of course continue to honor all of Anthropic’s non-disparagement and non-disclosure obligations, e.g. from mutual agreements.)
If you’ve signed one of these agreements and have concerns about it, please email hr@anthropic.com.
Hmm, I feel like you didn’t answer my question. Can you confirm that Anthropic has asked at least some employees to sign confidential non-disparagement agreements?
I think your previous comment pretty strongly implied that you think you did not do so (i.e. saying any previous agreements were merely “unclear” I think pretty clearly implies that none of them did include a non-ambiguous confidential non-disparagement agreement). I want to it to be confirmed and on the record that you did, so I am asking you to say so clearly.
“Unclear on this point” means what you think it means and is not a L I E for a spokesperson to say in my book. You got the W here already
I really think the above was meant to imply that the non disparagement agreements were merely unclear on whether they were covered by a non disclosure clause (and I would be happy to take bets on how a randomly selected reader would interpret it).
My best guess is Sam was genuinely confused on this and that there are non disparagement agreements with Anthropic that clearly are not covered by such clauses.
EDIT: Anthropic have kindly released me personally from my entire concealed non-disparagement, not just made a specific safety exception. Their position on other employees remains unclear, but I take this as a good sign
Thanks for this update! To clarify, are you saying that you WILL enforce existing non disparagements for everything apart from safety, but you are specifically making an exception for safety?
Given this part, I find this surprising. Surely if you think it’s bad to ask future employees to sign non disparagements you should also want to free past employees from them too?
This comment appears to respond to habryka, but doesn’t actually address what I took to be his two main points—that Anthropic was using NDAs to cover non-disparagement agreements, and that they were applying significant financial incentive to pressure employees into signing them.
Were these agreements subject to NDA? And were all departing employees asked to sign them, or just some? If the latter, what determined who was asked to sign?
I’m curious as to why it took you (and therefore Anthropic) so long to make it common knowledge (or even public knowledge) that Anthropic used non-disparagement contracts as a standard and was also planning to change its standard agreements.
The right time to reveal this was when the OpenAI non-disparagement news broke, not after Habryka connects the dots and builds social momentum for scrutiny of Anthropic.
I do want to be clear that a major issue is that Anthropic used non-disparagement agreements that were covered by non-disclosure agreements. I think that’s an additionally much more insidious thing to do, that contributed substantially to the harm caused by the OpenAI agreements, and I think is important fact to include here (and also makes the two situations even more analogous).
Note, since this is a new and unverified account, that Jack Clark (Anthropic co-founder) confirmed on Twitter that the parent comment is the official Anthropic position https://x.com/jackclarkSF/status/1808975582832832973
Thank you for responding! (I have more comments and questions but figured I would shoot off one quick question which is easy to ask)
Can you clarify what you mean by “even in these narrow cases”? If I am understanding you correctly, you are saying that you were including a non-disparagement clause by default in all of your severance agreements, which sounds like the opposite of narrow (edit: though as Robert points out it depends on what fraction of employees get offered any kind of severance, which might be most, or might be very few).
I agree that it would have technically been possible for you to also include such an agreement on start of employment, but that would have been very weird, and not even OpenAI did that.
I think using the sentence “even in these narrow cases” seems inappropriate given that (if I am understanding you correctly) all past employees were affected by these agreements. I think it would be good to clarify what fraction of past employees were actually offered these agreements.Severance agreements typically aren’t offered to all departing employees, but usually only those that are fired or laid off. We know that not all past employees were affected by these agreements, because Ivan claims to not have been offered such an agreement, and he left[1] in mid-2023, which was well before June 1st.
Presumably of his own volition, hence no offered severance agreement with non-disparagement clauses.
Ah, fair, that would definitely make the statement substantially more accurate.
@Sam McCandlish: Could you clarify whether severance agreements were also offered to voluntarily departing employees, and if so, under which conditions?
To expand on my “that’s a crux”: if the non-disparagement+NDA clauses are very standard, such that they were included in a first draft by an attorney without prompting and no employee ever pushed back, then I would think this was somewhat less bad.
It would still be somewhat bad, because Anthropic should be proactive about not making those kinds of mistakes. I am confused about what level of perfection to demand from Anthropic, considering the stakes.
And if non-disparagement is often used, but Anthropic leadership either specified its presence or its form, that would seem quite bad to me, because mistakes of commision here are more evidence of poor decisionmaking than mistakes of omission. If Anthropic leadership decided to keep the clause when a departing employee wanted to remove the clause, that would similarly seem quite bad to me.
I think that both these clauses are very standard in such agreements. Both severance letter templates I was given for my startup, one from a top-tier SV investor’s HR function and another from a top-tier SV law firm, had both clauses. When I asked Claude, it estimated 70-80% of startups would have a similar non-disparagement clause and 80-90% would have a similar confidentiality-of-this-agreement’s-terms clause. The three top Google hits for “severance agreement template” all included those clauses.
These generally aren’t malicious. Terminations get messy and departing employees often have a warped or incomplete picture of why they were terminated–it’s not a good idea to tell them all those details, because that adds liability, and some of those details are themselves confidential about other employees. Companies view the limitation of liability from release of various wrongful termination claims as part of the value they’re “purchasing” by offering severance–not because those claims would succeed, but because it’s expensive to explain in court why they’re justified. But the expenses disgruntled ex-employees can cause is not just legal, it’s also reputational. You usually don’t know which ex-employee will get salty and start telling their side of the story publicly, where you can’t easily respond with your side without opening up liability. Non-disparagement helps cover that side of it. And if you want to disparage the company, in a standard severance letter that doesn’t claw back vested equity, hey, you’re free to just not sign it–it’s likely only a bonus few weeks/months’ salary that you didn’t yet earn on the line, not the value of all the equity you had already vested. We shouldn’t conflate the OpenAI situation with Anthropic’s given the huge difference in stakes.
Confidentiality clauses are standard because they prevent other employees from learning the severance terms and potentially demanding similar treatment in potentially dissimilar situations, thus helping the company control costs and negotiations in future separations. They typically cover the entire agreement and are mostly about the financial severance terms. I imagine that departing employees who cared could’ve ask the company for a carve-out on the confidentiality for the non-disparagement clause as a very minor point of negotiation.
It’s great that Anthropic is taking steps to make these docs more departing-employee-friendly. I wouldn’t read too much into that the docs were like this in the first place (as this wasn’t on cultural radars until very recently) or that they weren’t immediately changed (legal stuff takes time and this was much smaller in scope than in the OpenAI case).
Example clauses in default severance letter from my law firm:
Are you able to elaborate on how this works? Are there any other details about this publicly, couldn’t find more detail via a quick search.
Some specific qs I’m curious about: (a) who handles the anonymous complaints, (b) what is the scope of behavior explicitly (and implicitly re: cultural norms) covered here, (c) handling situations where a report would deanonymize the reporter (or limit them to a small number of people)?
Anthropic has not published details. See discussion here. (I weakly wish they would; it’s not among my high-priority asks for them.)
OK, let’s imagine I had a concern about RSP noncompliance, and felt that I needed to use this mechanism.
(in reality I’d just post in whichever slack channel seemed most appropriate; this happens occasionally for “just wanted to check...” style concerns and I’m very confident we’d welcome graver reports too. Usually that’d be a public channel; for some compartmentalized stuff it might be a private channel and I’d DM the team lead if I didn’t have access. I think we have good norms and culture around explicitly raising safety concerns and taking them seriously.)
As I understand it, I’d:
Remember that we have such a mechanism and bet that there’s a shortcut link. Fail to remember the shortlink name (reports? violations?) and search the list of “rsp-” links; ah, it’s rsp-noncompliance. (just did this, and added a few aliases)
That lands me on the policy PDF, which explains in two pages the intended scope of the policy, who’s covered, the proceedure, etc. and contains a link to the third-party anonymous reporting platform. That link is publicly accessible, so I could e.g. make a report from a non-work device or even after leaving the company.
I write a report on that platform describing my concerns[1], optionally uploading documents etc. and get a random password so I can log in later to give updates, send and receive messages, etc.
The report by default goes to our Responsible Scaling Officer, currently Sam McCandlish. If I’m concerned about the RSO or don’t trust them to handle it, I can instead escalate to the Board of Directors (current DRI Daniella Amodei)
Investigation and resolution obviously depends on the details of the noncompliance concern.
There are other (pretty standard) escalation pathways for concerns about things that aren’t RSP noncompliance. There’s not much we can do about the “only one person could have made this report” problem beyond the included strong commitments to non-retaliation, but if anyone has suggestions I’d love to hear them.
I clicked through just now to the point of cursor-in-textbox, but not submitting a nuisance report.
Good that it’s clear who it goes to, though if I was an anthropic I’d want an option to escalate to a board member who isn’t Dario or Daniella, in case I had concerns related to the CEO
Makes sense—if I felt I had to use an anonymous mechanism, I can see how contacting Daniela about Dario might be uncomfortable. (Although to be clear I actually think that’d be fine, and I’d also have to think that Sam McCandlish as responsible scaling officer wouldn’t handle it)
If I was doing this today I guess I’d email another board member; and I’ll suggest that we add that as an escalation option.
Are there currently board members who are meaningfully separated in terms of incentive-alignment with Daniella or Dario? (I don’t know that it’s possible for you to answer in a way that’d really resolve my concerns, given what sort of information is possible to share. But, “is there an actual way to criticize Dario and/or Daniella in a way that will realistically be given a fair hearing by someone who, if appropriate, could take some kind of action” is a crux of mine)
Absent evidence to the contrary, for any organization one should assume board members were basically selected by the CEO. So hard to get assurance about true independence, but it seems good to at least to talk to someone who isn’t a family member/close friend.
(Jay Kreps was formally selected by the LTBT. I think Yasmin Razavi was selected by the Series C investors. It’s not clear how involved the leadership/Amodeis were in those selections. The three remaining members of the LTBT appear independent, at least on cursory inspection.)
I think that personal incentives is an unhelpful way to try and think about or predict board behavior (for Anthropic and in general), but you can find the current members of our board listed here.
For whom to criticize him/her/them about what? What kind of action are you imagining? For anything I can imagine actually coming up, I’d be personally comfortable raising it directly with either or both of them in person or in writing, and believe they’d give it a fair hearing as well as appropriate follow-up. There are also standard company mechanisms that many people might be more comfortable using (talk to your manager or someone responsible for that area; ask a maybe-anonymous question in various fora; etc). Ultimately executives are accountable to the board, which will be majority appointed by the long-term benefit trust from late this year.
Re 3 (and 1): yay.
If I was in charge of Anthropic I just wouldn’t use non-disparagement.
As a point of clarification: is it correct that the first quoted statement above should be read as “at least one employee” in line with the second quoted statement? (When I first read it, I parsed it as “all employees” which was very confusing since I carefully read my contract both before signing and a few days ago (before posting this comment) and I’m pretty sure there wasn’t anything like this in there.)
Current employees of OpenAI also wouldn’t yet have signed or even known about the non-disparagement agreement that is part of “general release” paperwork on leaving the company. So this is only evidence about some ways this could work at Anthropic, not others.
Yep, both should be read as “at least one employee”, sorry for the ambiguity in the language.
FWIW I recommend editing OP to clarify this.
Agreed, I think it’s quite confusing as is
Added a “at least some”, which I hope clarifies.
I am disappointed. Using nondisparagement agreements seems bad to me, especially if they’re covered by non-disclosure agreements, especially if you don’t announce that you might use this.
My ask-for-Anthropic now is to explain the contexts in which they have asked or might ask people to incur nondisparagement obligations, and if those are bad, release people and change policy accordingly. And even if nondisparagement obligations can be reasonable, I fail to imagine how non-disclosure obligations covering them could be reasonable, so I think Anthropic should at least do away with the no-disclosure-of-nondisparagement obligations.
Does anyone from Anthropic want to explicitly deny that they are under an agreement like this?
(I know the post talks about some and not necessarily all employees, but am still interested).
I left Anthropic in June 2023 and am not under any such agreement.
EDIT: nor was any such agreement or incentive offered to me.
Neither is Daniel Kokotajlo. Context and wording strongly suggest that what you mean is that you weren’t ever offered paperwork with such an agreement and incentives to sign it, but there remains a slight ambiguity on this crucial detail.
Correct, I was not offered such paperwork nor any incentives to sign it. Edited my post to include this.
I am a current Anthropic employee, and I am not under any such agreement, nor has any such agreement ever been offered to me.
If asked to sign a self-concealing NDA or non-disparagement agreement, I would refuse.
Did you see Sam’s comment?
Agreed. I’d be especially interested to hear this from people who have left Anthropic.
I agree that this kind of legal contract is bad, and Anthropic should do better. I think there are a number of aggrevating factors which made the OpenAI situation extrodinarily bad, and I’m not sure how much these might obtain regarding Anthropic (at least one comment from another departing employee about not being offered this kind of contract suggest the practice is less widespread).
-amount of money at stake
-taking money, equity or other things the employee believed they already owned if the employee doesn’t sign the contract, vs. offering them something new (IANAL but in some cases, this could be a felony “grand theft wages” under California law if a threat to withhold wages for not signing a contract is actually carried out, what kinds of equity count as wages would be a complex legal question)
-is this offered to everyone, or only under circumstances where there’s a reasonable justification?
-is this only offered when someone is fired or also when someone resigns?
-to what degree are the policies of offering contracts concealed from employees?
-if someone asks to obtain legal advice and/or negotiate before signing, does the company allow this?
-if this becomes public, does the company try to deflect/minimize/only address issues that are made publically, or do they fix the whole situation?
-is this close to “standard practice” (which doesn’t make it right, but makes it at least seem less deliberately malicious), or is it worse than standard practice?
-are there carveouts that reduce the scope of the non-disparagement clause (explicitly allow some kinds of speech, overriding the non-disparagement)?
-are there substantive concerns that the employee has at the time of signing the contract, that the agreement would prevent discussing?
-are there other ways the company could retaliate against an employee/departing employee who challenges the legality of contract?
I think with termination agreements on being fired there’s often 1. some amount of severance offered 2. a clause that says “the terms and monetary amounts of this agreement are confidential” or similar. I don’t know how often this also includes non-disparagement. I expect that most non-disparagement agreements don’t have a term or limits on what is covered.
I think a steelman of this kind of contract is: Suppose you fire someone, believe you have good reasons to fire them, and you think that them loudly talking about how it was unfair that you fired them would unfairly harm your company’s reputation. Then it seems somewhat reasonable to offer someone money in exchange for “don’t complain about being fired”. The person who was fired can then decide whether talking about it is worth more than the money being offered.
However, you could accomplish this with a much more limited contract, ideally one that lets you disclose “I signed a legal agreement in exchange for money to not complain about being fired”, and doesn’t cover cases where “years later, you decide the company is doing the wrong thing based on public information and want to talk about that publically” or similar.
I think it is not in the nature of most corporate lawyers to think about “is this agreement giving me too much power?” and most employees facing such an agreement just sign it without considering negotiating or challenging the terms.
For any future employer, I will ask about their policies for termination contracts before I join (as this is when you have the most leverage, if they give you an offer they want to convince you to join).
This is true. I signed a concealed non-disparagement when I left Anthropic in mid 2022. I don’t have clear evidence this happened to anyone else (but that’s not strong evidence of absence). More details here
EDIT: I should also clarify that I personally don’t think Anthropic acted that badly, and recommend reading about what actually happened before forming judgements. I do not think I am the person referred to in Habryka’s comment.
In the case of OpenAI most of the debate was about ex-employees. Are we talking about current employees or ex-employees here?
I am including both in this reference class (i.e. when I say employee above, it refers to both present employees and employees who left at some point). I am intentionally being broad here to preserve more anonymity of my sources.
Not sure how to interpret the “agree” votes on this comment. If someone is able to share that they agree with the core claim because of object-level evidence, I am interested. (Rather than agreeing with the claim that this state of affairs is “quite sad”.)
A LOT depends on the details of WHEN the employees make the agreement, and the specifics of duration and remedy, and the (much harder to know) the apparent willingness to enforce on edge cases.
“significant financial incentive to promise” is hugely different from “significant financial loss for choosing not to promise”. MANY companies have such things in their contracts, and they’re a condition of employment. And they’re pretty rarely enforced. That’s a pretty significant incentive, but it’s prior to investment, so it’s nowhere near as bad.
A pre-existing market on this question https://manifold.markets/causal_agency/does-anthropic-routinely-require-ex?r=SmFjb2JQZmF1
What’s your median-guess for the number of times Anthropic has done this?
(Not answering this question since I think it would leak too many bits on confidential stuff. In general I will be a bit hesitant to answer detailed questions on this, or I might take a long while to think about what to say before I answer, which I recognize is annoying, but I think is the right tradeoff in this situation)
I’m kind of concerned about the ethics of someone signing a contract and then breaking it to anonymously report what’s going on (if that’s what your private source did). I think there’s value from people being able to trust each others’ promises about keeping secrets, and as much as I’m opposed to Anthropic’s activities, I’d nevertheless like to preserve a norm of not breaking promises.
Can you confirm or deny whether your private information comes from someone who was under a contract not to give you that private information? (I completely understand if the answer is no.)
(Not going to answer this question for confidentiality/glommarization reasons)
I think this is a reasonable question to ask. I will note that in this case, if your guess is right about what happened, the breaking of the agreement is something that it turned out the counterparty endorsed, or at least, after the counterparty became aware of the agreement, they immediately lifted it.
I still think there’s something to maintaining all agreements regardless of context, but I do genuinely think it matters here if you (accurately) expect the entity you’ve made the secret agreement with would likely retract it if they found out about it.
(Disclaimer that I have no private info about this specific situation.)