> *Isn’t it unreasonable for Anthropic to suddenly set terms in their contract?* The terms were in the original contract, which the Pentagon agreed to. It’s the Pentagon who’s trying to break the original contract and unilaterally change the terms, not Anthropic.
> *Doesn’t the Pentagon have a right to sign or not sign any contract they choose?* Yes. Anthropic is the one saying that the Pentagon shouldn’t work with them if it doesn’t want to. The Pentagon is the one trying to force Anthropic to sign the new contract.
[1]: https://www.astralcodexten.com/p/the-pentagon-threatens-anth...
> [1] This story requires some reading between the lines - the exact text of the contract isn’t available - but something like it is suggested by the way both sides have been presenting the negotiations.
I deal with far too many people who won't believe me without 10 bullet-proof sources but get very angry with me if I won't take their word without a source :(
> "Two such use cases have never been included in our contracts with the Department of War..."
While I agree with Anthropic's position on this regardless, the original contract wording does matter in terms of making either the government look even more unreasonable or Anthropic look a little less reasonable.
The issue is a subtle ambiguity in Dario's statement: "...have never been included in our contracts" because it leaves two possibilities: 1. those two conditions were explicitly mentioned and disallowed in the contract, or 2. they weren't in the contract itself - and are disallowed by Anthropic's Terms of Service and complying with the ToS is a condition in the contract (which would be typical).
If that's the case, then it matters if the ToS disallowed those two uses at the time the original contract was signed, or if the ToS was revised since signing. Anthropic is still 100% in the right if the ToS disallowed these uses at the time of signing and the ToS was an explicit condition of the contract, since contracts often loop in the ToS as a condition while not precluding the ToS being updated.
However, if the ToS was updated after contract signing and Anthropic added or expanded the wording of those two provisions, then the DoD, IMHO, has a tiny shred of justification to complain and stop using Anthropic. Of course, going much further and banning the entire US government (and contractors) from using Anthropic for any use, including all the ones where these two provisions don't matter - is egregiously punitive and shitty.
While the contract wording itself may be subject to NDA, it would be helpful if Anthropic's statements could be a bit more precise. For example, if Dario had said "have always been disallowed in our contracts" this ambiguity wouldn't exist.
You're trying desperately to find a way that things can be at least a little normal, and I really do get it. It would be great if such a way existed. But it doesn't. I recommend you take a social media break like I'm about to, take the time you need to mourn the era of normal politics, and come back with a full understanding that the US government is not pursuing normal policy objectives with bad decisions. They hate you and they hate me for not being on their side, and their primary goal is to ensure that we're as miserable as they can make us.
Disregarding who is right or wrong for a moment, if the DoW are right (which I'm not personally inclined to believe, but we're ignoring that for the moment) -- how else can they avoid secondhand Claude poisoning?
Supposing they really want to use their software for things disallowed by Claude's (now or future) ToS, it seems like designating it a supply chain risk is the only way they can ensure that their contractors don't include Claude (either indirectly as a wrapper or tertially through use of generated code etc)
I agree that if the DoW claim is correct (and I doubt it is), then, sure, the DoW dropping Anthropic and precluding the DoW's suppliers from using Anthropic for any DoW work would be expected. However, the "supply chain risk" designation they are deploying goes far beyond that to block Anthropic use by any supplier to any part of the entire U.S. government for anything.
For example, no one at Crayola can use Anthropic for anything because Crayola sells crayons to the Education Dept. The DoW already has much less draconian ways to restrict what their direct suppliers use to build things for military applications. But instead of addressing the actual risk in a normal measured way, they are choosing to use a nuke against a grenade-sized problem. This "supply chain risk" designation is rarely used and has never been used against a U.S. company. It's used against Chinese or Russian companies when in cases where there's credible risk of sabotage or espionage. That's why that particular designation always blocks all products from an entire company for any application by any part of the U.S. Government, contractors and suppliers (which is why it's never been used against a U.S. company).
My hope is that this gets us some real concern for things that have been defended with de facto arguments (i.e. privacy) going forward.
edit: Anthropic argues that your Crayola analogy is fundamentally incorrect.
> Legally, a supply chain risk designation under 10 USC 3252 can only extend to the use of Claude as part of Department of War contracts—it cannot affect how contractors use Claude to serve other customers.
https://www.anthropic.com/news/statement-comments-secretary-...
Yes, I just saw Dario's latest post with that more detailed info. My understanding was informed by news reporting in a couple different outlets but those reports may have been conflating the "supply chain risk" designation (under 10 USC 3252) with the net effect of statements from the pentagon and white house which go substantially further.
Even if it's not in the legal scope of 10 USC 3252, the administration has made clear they intend to ban Anthropic from use across the federal government. AFAICT doing that is probably within the discretionary remit of the executive branch, even though I believe it's unprecedented - to your point about de jure and de facto law.
To me, if there's a silver lining to all this, it's making a strong case for restricting executive branch power.
Edit to add: Per the Wall Street Journal's lead story (updated in the last hour): "The General Services Administration, which oversees federal procurement, said it is removing Anthropic from its product offerings to government agencies... Even absent the supply-chain risk designation, broadening the clash to include all federal agencies takes the Anthropic fight to a much larger scale than its spat with the Pentagon."
Because the act that the SCR designation would “protect” against is treason, so I don’t think people would care too much whether there’s a contract.
> The Leftwing nut jobs at Anthropic have made a DISASTROUS MISTAKE trying to STRONG-ARM the Department of War, and force them to obey their Terms of Service instead of our Constitution.
(Or maybe it’s catchier to respond glibly with “never trust a child rapist and convicted felon.”)
The Pentagon feels it isn't Anthropic to set boundaries as to how their tech is used (for defense) since it can't force its will, then it bans doing business with them.
Once the hive decides you're being serious without checking, they turn the down vote button into an I disagree with you button.
This is actually one of the reasons I left Reddit. I hate to see it here.
The DoD declaration reflects a certain context, we had the patriotic act, a whistleblower exiled in Russia for defending the constitution, etc etc. We didn't need to wait a MAGA movement to be expecting such comment from the DoD.
If hackernews threads turn into mouthpieces for opinions then we have no use posting anything in here.
The comments are naively claiming commercial agreements make Anthropic right, as if contracts had more weight than the constitution.
I would rather call out a "virtuous signalling" entity in the valley simply standing for something aligned with civil liberties, and using it as a political stance in what nobody would deny is an unfortunate polarized political climate.
What to make of OpenAI then. Should I give my opinion that they took a falsely constitutional stance, or simply made for-profit move to land a juicy government contract, while making the public think they kept the same red lines as their main competitor?
Or just stick to the fact: The DoD will, as always, get away with its liberticide demands to get what it wants, because other big tech will fall inline.
I may not agree with what people say and it seems like he may have just been kidding or was being sarcastic, but he should be allowed to say it without being bullied and abused by downvotes.
I hope everyone will reconsider their ways.
So it's six months to phase out use of Anthropic at the DoD, but the people hosting the models have to stop "immediately".
Which miiight impact the amount of inference the DoD would be able to get done in those six months.
> Which miiight impact the amount of inference the DoD would be able to get done in those six months.
Which might not be by accident looking at the Truth Social posts which state "Anthropic better get their act together, and be helpful during this phase out period, or I will use the Full Power of the Presidency to make them comply, with major civil and criminal consequences to follow."
I would not be surprised to see this being used as an excuse to nationalize Anthropic.
According to whom?
No, not this room. The one with Hegseth in it.
Look at his other comments. He's not wrong.
The US isn't Iran, North Korea, or even China, as much as some people, including the US president, seem want to emulate those models.
Imagine a _leaded_ pipe supplier not being allowed to tell the department of war they shouldn't use leaded pipes for drinking water! It's the job of the vendor to tell the customer appropriate usage.
Go look at the package on a kitchen knife and it says not to be used as a weapon
I'm not sure this is still a useful analogy, though...
The knife manufacturer isn't obligated to sell to you in either case, I'd expect them not to cut ties with you in the self defence scenario. But it is their choice.
1. Found out you used their knives to go murdering
2. Sells knives in a fashion where it's possible for them to prevent you from buying their knives (i.e. direct to consumer sales)
Would almost certainly not "be more than happy to continue to sell to you". Even if we ignore the fact that most people are simply against assisting in murders (which by itself is a sufficient justification in most companies), the bad PR (see the "found out" and "direct to consumer" part) would make you a hugely unprofitable customer.
> the bad PR (see the "found out" and "direct to consumer" part) would make you a hugely unprofitable customer.
That... Doesn't happen.
Boycotts by people who weren't going to buy your product anyway are immaterial to business. The inevitable lawsuits are costly, but are generally thought of as good publicity, because they keep the business name in the news.
People who buy (and make) firearms are... pretty close to the exact opposite.
Goalposts moved.
Where either it's a computer program (website) that knows nothing about you, or cutco.
If you think you wouldn't find a cutco representative to sell to you, you're on some good reality-altering drugs.
There will always be some company willing to sell to even the worst person, in any product category.
The response that companies have to boycotts, and the results of the boycotts themselves, are fractally chaotic at best.
But even most nominally socially-aware companies are reactive, rather than proactive.
It's about how we think they will act.
Especially when it comes to sales to the US military, I have no expectations about how companies will act.
Hell, just look at how many companies willingly helped China with their Great Firewall.
What I mean is that you _did_ judge them by a standard used for weapons manufacturers. How you react to their actions _is_ your judgement.
But perhaps that is the standard we should use. Weapons manufacturing is a well regulated industry after all. Export controls, dual-use technology restrictions, if it has applications for warfare it should be appropriately restricted.
I think any of these companies will attempt to get away with whatever the fuck they can.
That has fuckall to do with your rhetorical question of:
> That's the standard we should judge them by?
Also, in the cases you can't, it's generally the government stopping you, not the gun companies.
Edit: hell I get downvoted and look where the knife analogy got us. A load of weird replies miles away from anything related to AI or DoD.
Claude Opus is just remarkably good at analysis IMO, much better than any competitor I’ve tried. It was remarkably good and complete at helping me with some health issues I’ve had in the past few months. If you were to turn that kind of analytical power in a way to observe the behaviour of American citizens and to change it perhaps, to make them vote a certain way. Or something like - finding terrorists, finding patterns that help you identify undocumented people.
Within a few iterations of this it gave me a simple blood panel, then I did that one and it kept suggesting more simple lab or at home tests and we kept going through them until I was reasonably certain of “something” and now that I have hypothesis I am going to a doctor. I think it’s done a great job. I also kept asking it for simple lifestyle interventions to prevent progression of my issue and it consistent nailed it - one particular interverntion (adding salt to water and drinking it to prevent symptoms) made a huge improvement to my life - I was barely working before that.
I added in some text the instructions box (project master prompt) for it to realise - it’s not medical advice and I am aware of that (prevents excessive guardrails) - add confidence intervals and probability to all diagnostic statements (prevents me + Claude going into rabbit holes so easily, it often has 70-80% certainty of what it’s saying, but it’s clear that it doesn’t use the right language) - that It was talking to an non expert, to use simple language but to go into detail when necessary. I also ask it to stop doing unnecessary constant follow up questions to every answer as that causes me anxiety. I can share the prompt, in fact I might do so later as it might be useful to others.
Make sure your first chat is about the exams in the project files. Make sure it reads them all. It has a tendency to read a few and go “is this good”. Ask for a summary and note any absences.
Try using the research and extended thinking features a lot if you think it’s not fully aware of anything. It might not be aware of more recent research. If it’s a serious condition you are researching, just ask it to do sweeps / use research to look for new info about it and find new papers. It might also deepen its understanding.
After you do research you can make a simple artefact and throw it onto the project files. That allows it to refer to it and gain more knowledge about a condition or issue that might not be as rich in the training data.
So, I find GPT to be so so bad for this it made me realise a bit on why the USG is so insistent. Claude Opus is just on a different class.
Here’s the master project prompt:
Act as an expert who’s talking to an interested layman. Engage in detail when requested but be overall succinct in your answers. Short sentences are fine, no need into be lengthy. Do deep research. When arriving at any kind of conclusion or hypothesis assign it a probability and a confidence interval - define this in percentages as in “90%”
On Artefacts - all artefacts should be just text and markdown. Never do anything more complicated with formatting, unless by explicit request.
Don't ask follow up questions unless it's to make for better diagnosis. I.e. don't keep asking questions just to maintain conversation going please. But never hesitate to ask questions if it makes for better outcomes.
What we're seeing here is that when a vendor declines to change the terms of its contractual agreement for ethical reasons, the government publicly attacks it.
With the other stated reason being legal. "To the extent that such surveillance is currently legal, this is only because the law has not yet caught up with the rapidly growing capabilities of AI."
I don't think we should lessen Anthrophic's stance from technical/legal to ethical. Just as we shouldn't describe what the department of war is doing as "not renewing a contract".
Congress could come up with an act it it's for national interest.
The military isn't the typical End User.
Utter nonsense. When the US built the Blackbird, it could only use titanium because of the heat involved in traveling at that speed. But they didn't have enough titanium in the US. So the the US created front companies to purchase titanium from the Soviet Union.
Do you think the US should have informed the Soviet Union what it wanted to do with the metal?
Your comparison seems backwards
If this were a news outline writing "Department of War" I would be concerned. But in the case of the Anthropic CEO's blog post, I can understand why they are picking their fights.
Sad.
It's a testament to the broken information ecosystem we're in that many people genuinely don't know this. Most will correct themselves when told. I agree with you that those who don't are not worth engaging.
Depends where you at
(Spoiler: it's still legally called the Gulf of America)
They want the department to fight wars. At least they’re being honest.
Declaring a war requires Congress, and fighting a war other than in response to an invasion may be illegal under US law if Congress has not exercised its power to declare war, but that doesn't prevent wars from happening it just makes it illegal (though the only actual remedy is impeachment) for the President to wage war without authorization. And, in any case, that’s largely moot because Congress has exercised that power in an open ended (in terms of when and against whom) but limited (in authorized duration of any particular action without subsequent authorization) manner via the War Powers Act, giving every President since Nixon a blank check to start wars with full legal authority and then allow Congress an opportunity to vote to pull support from forces already in combat and hope the enemy already engaged is willing to treat the war as over as the only after-the-fact constraint.
I can't see anyway this ends well for the US. I say this as both an American and a military veteran.
Frederick VII of Denmark, an absolute monarch, introduced parliamentarism without any violence or even broad public pressure.
And thats just what I can remember without digging.
[1] "only an act of Congress can formally change the name of a federal department." https://en.wikipedia.org/wiki/Executive_Order_14347
(edited to add the url I omitted)
"You can just do things" (evil edition).
Dario's framing: "we said no to the bad stuff, we only do the good stuff", is textbook damage control. He's not disputing that Anthropic works with the DoW. He's trying to control the narrative around how they work with the DoW before critics on both sides define it for him. The carefully drawn line between "acceptable" and "unacceptable" military use cases reads less like a principled stand and more like a retroactive justification designed to thread the needle between defense revenue and public image.
Ask yourself: if the directive had never been made public, would that blog post exist? Almost certainly not. Anthropic would still be quietly serving the DoW, and none of us would be having this conversation.
That doesn't necessarily make the DoW right or Anthropic wrong on the contract dispute itself. But treating Dario's statement as a neutral, authoritative account of what happened, rather than what it obviously is, which is a company protecting its brand in a crisis, is doing a lot of the DoW's work for them by accepting one side's PR as fact.
A fact being technically available and that fact commanding widespread public attention are very different things. Anthropic's communications team understands this distinction even if you don't find it interesting. The blog post wasn't written for people who already track federal AI contracts, it was written for the much larger audience encountering this story for the first time and forming opinions about it in real time.
If the point you're making is just "I already knew this," that's fine, but it doesn't address anything about the incentive structure behind the public response.
You're ignoring the sequence of events on the ground.
If there hadn't been any been any internal pushback from Anthropic, would the directive have ever been made public?
Internal pushback and public damage control aren't mutually exclusive. A company can genuinely disagree with a client's demands behind closed doors and simultaneously craft a public narrative designed to make itself look as good as possible once those disagreements surface. In fact, that's exactly what competent communications teams do, they plan for the scenario where private disputes become public, and they have messaging ready.
The real question isn't who went public first or why. It's whether Anthropic's stated position, "we support these military use cases but not those ones", reflects a durable ethical framework or a line drawn precisely where it needed to be to keep both the contracts and the brand intact. Nothing in the sequencing you've described answers that question. It just tells us Anthropic saw this coming, which, if anything, means the messaging was more carefully engineered, not less.
[Edit: update]
I use assistive voice transcription because I'm unable to type well with a keyboard. But I'd point out that "you must be an AI" has become the new way to dismiss an argument without engaging with it. It's the modern equivalent of "you're just copy-pasting talking points", it lets you discard everything someone said without addressing a single word of it.
The fact that my sentences "thread together" is not evidence of anything other than coherent thinking. And speed of response says more about the tools someone uses than whether a human is behind them. Plenty of people use dictation, accessibility tools, or just happen to type fast.
^^^ This took me 30 seconds to speak aloud.
They will just have to recompete!
Trump has historically stiffed his contractors. Why do you think his administration would be any different with adhering to a contract?
My speculation is the "business records" domestic surveillance loophole Bush expanded (and that Palantir is build to service). That's usually how the government double-speaks its very real domestic surveillance programs. "It's technically not the government spying on you, it's private companies!" It's also why Hegseth can claim Anthropic is lying. It's not about direct government contracts. It's about contractors and the business records funnel.
Of course they can just say - we aren’t, Palantir is.
The basic problem in our polity is that we've collectively transferred the guilty pleasure of aligning a charismatic villain in fiction to doing the same in real life. The top echelons of our government are occupied by celebrities and influencers whose expertise is in performance rather than policy. For years now they've leaned into the aesthetics of being bad guys, performative cruelty, committing fictional atrocities, and so forth. Some MAGA influencers have even adopted the Imperial iconography from Star Wars as a means of differentiating themselves from liberal/democratic adoption of the 'rebel' iconography. So you have have influencers like conservative entrepreneur Alex Muse who styles his online presence as an Imperial stormtrooper. As Poe's law observes, at some point the ironic/sarcastic frame becomes obsolete and you get political proxies and members of the administration arguing for actual infringements of civil liberties, war crimes, violations of the Constitution and so on.
Like, Mark Hamill himself is a massive Israel + Biden supporter [0].
Guys, George Lucas didn't make the Empire thinking about Trump, or Republicans. He made it about America.
0 - https://www.nme.com/news/film/hollywood-stars-sign-open-lett...
[1] https://xcancel.com/MarkHamill/status/1725979647991537786?la...
So one thing to call out here is that the assumption that DoW is working on specifically these use cases is not bullet proof. They simply may not want to share with anthropic exactly what they are working on for natsec issues. /we can't tell you/ could violate the terms.
It is also dumb that DoW accepted these terms in the first place.
Look, Anthropic is not going to be designated a supply chain risk. 80% of the Fortune 500 have contracts with them. Probably a similar percentage of defense contractors. Amazon is a defense contractor for example. They'd have to remove Claude from their AWS offerings. Everyone running Claude on AWS, boom gone. The level of disruption to the US economy would be off the charts, and for what? Why? Because Hegseth had a bad day? Because he's a sore loser?
If he's decided he doesn't like the DoW's contract then he can cancel it, fine. To try and exact revenge on the best American frontier model along with 80% of the Fortune 500 in the process, to go out of his way to harm hundreds or perhaps thousands of American firms, defies all reason. This is behavior you would expect any adult would understand as petty and foolish, let alone one who's made it to the highest ranks of government.
So I think it's just not going to happen, Trump's statement on the matter notably didn't mention a supply chain risk designation. This suggests to me that Hegseth went off half cocked. The guy is a liability for Trump at this point, I'm guessing he won't last much longer.
seriously? :)
No doubt the US Gov't will be using A I to perform automated military strikes without human supervision. and spying on US citizens (which they already have been doing for decades now).
Look no further than the case of patriot Mark Klein, a former AT&T technician, exposed a massive NSA surveillance program in 2006, revealing that AT&T allowed the government to intercept, copy, and monitor massive amounts of American internet traffic. Klein discovered a secret, NSA-controlled room—Room 641A—inside an AT&T facility in San Francisco, which acted as a splitter for internet traffic.
Anthropic wouldn't have walked away from a multi million contract if their two redlines could be respected. OpenAI on the other hand is a fast, willing and ready company. I would love to see Anthropic's proposed contract
In our agreement, we protect our red lines through a more expansive, multi-layered approach. We retain full discretion over our safety stack, we deploy via cloud, cleared OpenAI personnel are in the loop, and we have strong contractual protections. This is all in addition to the strong existing protections in U.S. law.
We believe strongly in democracy. Given the importance of this technology, we believe that the only good path forward requires deep collaboration between AI efforts and the democratic process. We also believe our technology is going to introduce new risks in the world, and we want the people defending the United States to have the best tools.
Our agreement includes:
1. Deployment architecture. This is a cloud-only deployment, with a safety stack that we run that includes these principles and others. We are not providing the DoW with “guardrails off” or non-safety trained models, nor are we deploying our models on edge devices (where there could be a possibility of usage for autonomous lethal weapons).
Our deployment architecture will enable us to independently verify that these red lines are not crossed, including running and updating classifiers.
2. Our contract. Here is the relevant language:
The Department of War may use the AI System for all lawful purposes, consistent with applicable law, operational requirements, and well-established safety and oversight protocols. The AI System will not be used to independently direct autonomous weapons in any case where law, regulation, or Department policy requires human control, nor will it be used to assume other high-stakes decisions that require approval by a human decisionmaker under the same authorities. Per DoD Directive 3000.09 (dtd 25 January 2023), any use of AI in autonomous and semi-autonomous systems must undergo rigorous verification, validation, and testing to ensure they perform as intended in realistic environments before deployment.
For intelligence activities, any handling of private information will comply with the Fourth Amendment, the National Security Act of 1947 and the Foreign Intelligence and Surveillance Act of 1978, Executive Order 12333, and applicable DoD directives requiring a defined foreign intelligence purpose. The AI System shall not be used for unconstrained monitoring of U.S. persons’ private information as consistent with these authorities. The system shall also not be used for domestic law-enforcement activities except as permitted by the Posse Comitatus Act and other applicable law.
I would assume the original terms the DoW is now railing against were in those original contracts that they signed. In that case it looks like the DoW is acting in bad faith here, they signed the original contact and agreed to those terms, then they went back and said no, you need to remove those safeguards to which Anthropic is (rightly so) saying no.
Am I missing something here?
EDIT: Re-reading Dario's post[1] from this morning I'm not missing anything. Those use cases were never part of the original contacts:
> Two such use cases have never been included in our contracts with the Department of War
So yeah this seems pretty cut and dry. Dow signed a contract with Anthropic and agreed to those terms. Then they decided to go back and renege on those original terms to which Anthropic said no. Then they promptly threw a temper tantrum on social media and designated them as a supply chain risk as retaliation.
My final opinion on this is Dario and Anthropic is in the right and the DoW is acting in bad faith by trying to alter the terms of their original contracts. And this doesn't even take into consideration the moral and ethical implications.
[1]: https://www.anthropic.com/news/statement-department-of-war