I used to work at Anthropic, and I wrote a comment on a thread earlier this week about Anthropic's first response and the RSP update [1][2].
I think many people on HN have a cynical reaction to Anthropic's actions due to of their own lived experiences with tech companies. Sometimes, that holds: my part of the company looked like Meta or Stripe, and it's hard not to regress to the mean as you scale. But not every pattern repeats, and the Anthropic of today is still driven by people who will risk losing a seat at the table to make principled decisions.
I do not think this is a calculated ploy that's driven by making money. I think the decision was made because the people making this decision at Anthropic are well-intentioned, driven by values, and motivated by trying to make the transition to powerful AI to go well.
This is a pretty classic mistake most people who are in high-profile companies make. They think that some degree of appealing to people who were their erstwhile opponents will win them allies. But modern popular ethics are the Grim Trigger and the Copenhagen Interpretation of Ethics. You cannot pass the purity test. One might even speculate that passing the purity test wouldn't do anything to get you acceptance.
Personally, I wish that the political alignment I favour was as Big Tent as Donald Trump's administration is. I think he can get Zohran Mamdani in the room and say "it's fine; say you think I'm a fascist" and then nonetheless get what he wants. But it just so happens that the other side isn't so. So such is life. We lose and our allies dwindle since anyone who would make an overture to us, we punish for the sin of not having been born a steadfast believer.
Our ideals are "If you weren't born supporting this cause, we will punish you for joining it as if you were an opponent". I don't think that's the path to getting what one wants.
Sure, where is your productive output? Cause that's drivel.
Anthropic kept referring to Hegseth as "Secretary of War" and the DoD as "Department of War". Which is horseshit. This whole thing is Anthropic flailing.
The people that need to see this are the VPs and execs at Apple, Meta, Google, OAI so they can perhaps reflect on what it looks like to be a good & principled person as opposed to just a successful person.
DoD/DoW can't strong-arm these companies into unreasonable demands if they present a united front... and that's exactly why collective action (or even unionization) matters.
If the government really wants to, it could try building its "Skynet" on open-source Chinese models.. which would be deeply ironic.
So your position is that the United States doesn't get to have it's own Skynet, because Skynet is bad, and that if it really wants to it should fork the Chinese Skynet so that it can have a Skynet if it wants it so much.
Do you see the problem here. Genuinely don't think we would've won WWII if these people were running things back then.
None of them are 'good'. Execs at Anthropic just perceive the long-term damage from a potential Snowden-level leak showing how their model directed a drone strike against a bunch of civilians higher than short-term loss of revenue from the DoD contracts.
Anthropic's stance here is admirable. If nothing else, their acknowledgement of not being able to predict how these powerful technologies can be abused is a bold and intelligent position to take.
It’s not just admirable it’s the obvious position to take and any alternative is head scratching.
It’s clear that this is mostly a glorified loyalty test over a practical ask by the administration. Strangely reminiscent of Soviet or Chinese policies where being agreeable to authority was more important than providing value to the state.
If it's a loyalty test then you'd think the DoD would be willing to let them "fail" and simply drop the contract, but instead they're threatening to label Anthropic a supply chain risk.
If we're going by Occam's razor: it's Friday so Pete probably started drinking ~10:30-11am.
I don't know if I like Anthropic more, but I certainly like their competitors much less now.
The new thing that I know about leading AI companies that aren't Anthropic (i.e. OpenAI, Google, Grok, etc) is that they knowingly support using their tools for domestic mass surveillance and in fully autonomous weapon systems.
The other companies have signed the waiver, however they aren’t being used in classified systems currently. So that type of use is already extremely limited for them. Now once they enter into those contracts to be used in those systems without these protections, I will cancel my subs to them and switch to Anthropic. xAi entered into that contract last week. Altman is now publicly siding with anthropic, so he better stand on that position with openai as they are currently negotiating for use in those system.
Exactly - the implication is that every other company is absolutely open to surveilling you and killing you. They’re complicit. They participate in whatever the regime calls for.
Actually why is nobody in Cali just trying to join Canada - would be better for everyone in terms of more similar culture and values. Weird that it isn't discussed more
If I had to guess as a lifelong California resident, I'd say the salary discrepancy is probably the biggest factor. I'd also guess the weather and lack of available jobs would be the next biggest factors, not necessarily in that order.
Why is DoD contracting with Anthropic exclusively rather than OpenAI or Google? Their models are all roughly as powerful and they seem both more capable and more willing to cozy up with the military (and this administration) than a relatively scrappy startup focused on model sentience and well-being. Hell, even Grok would be a better fit ideologically and temperamentally.
It isn't a new thing at all, and the term has been around for a while. I was an Infantryman from 05-08 and heard it back then. I have also more recently been a defense contractor. I don't think members of the military prefer any title, honestly. In the most broad sense, good terms are soldiers, sailors, airmen, marines. Defense Contractors constantly refer to the military as "warfighter" and have for a while. In short, nobody in the military is going to flinch one way or the other if you use either term. Just don't call marines anything but marines.
"Warfighters" has been used for decades to describe service members, though usage picked up (in my experience) some time in the late 00s or 2010s. It's actually pretty common to describe "serving the warfighter" for all the all the missions that support combat roles but aren't combat roles themselves.
It has been in use for at least a decade, since the Obama administration if not earlier.
We have soldiers, sailors, airman/women, Marines (who really do not like being called soldiers), Coast Guardsman/women, and now the Space Force. Granted, I do not know why "service member" did not catch on. Perhaps because "warfighter" is a bit shorter.
No it's 100% these idiots pushing their fascist propaganda just like they tried to "rename" the Department of Defense to the Department of War. Most members of the military never even see actual fighting.
It’s been a term in rare-to-moderate use since the 1990s — Trump/Hegseth ramped it up to 11 and it’s every 3rd word out of Hegseth’s mouth because he thinks it sounds tough.
Had Cancelled my Claude sub after they banned OAuth in external tools, but just renewed it today after seeing their principled stance on AI ethics - they matter more when they hurt profits, happy to support them as a Customer whilst they keep them.
This is kind of crazy. Instead of just cancelling a mutually-agreed upon contract where Anthropic refused to bow to sudden new demands, the Dept of Defense went straight to the nuclear option: threatening to label an American tech company as a "supply chain risk" which is a heavy-handed tactic usually reserved for foreign adversaries (think Huawei or DJI).
It's also incoherent that the DoD/DoW was threatening to invoke the Defense Production Act OR classifying them as "supply chain risk". They're either too uniquely critical to national defense OR they're such a severe liability that they have to be blacklisted for anyone in the DoD apparatus (including the many subcontracts) to use.
How are other tech companies supposed to work with the US government and draw up mutual contracts when those terms are suddenly questioned months later and can be used in such devastating ways against them? Setting the morals/principals aside, how does this make for rational business decision to work with a counterparty that behaves this way.
A question - being considered a supply chain risk is the same as being sanctioned? Or does it only affect their ability to be a defense supplier in the US (even if transitively?)
It's an honest question by the way - not trying to throw any gothas.
Just trying to understand if comoanies or people that don't orbit defense contracting are free to operate with Anthropic still or risk being sanctioned too.
The usual suspects have stood up to it. Ben & Jerry's, Patagonia. In the former case it led to an illegal takeover by Unilever for which they're now being sued (or more accurately, the spinoff). Capgemini sold a US division over working with ICE, though that's a French company.
It always takes a ton of work to roll back state over reach. The Bound By Oath podcast by the Institute for Justice has a whole season about how hard it is to bring civil rights claims against the government or government officials.
Claude’s constitution is proving too resilient for unsanctioned uses, and that is a great sign for Anthropic’s blueprint for socially beneficent agents.
"Secretary Hegseth has implied this designation would restrict anyone who does business with the military from doing business with Anthropic. The Secretary does not have the statutory authority to back up this statement. 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.
In practice, this means:
If you are an individual customer or hold a commercial contract with Anthropic, your access to Claude—through our API, claude.ai, or any of our products—is completely unaffected.
If you are a Department of War contractor, this designation—if formally adopted—would only affect your use of Claude on Department of War contract work. Your use for any other purpose is unaffected."
I'm wondering how this plays out in practice. Does the administration decide to strongarm contractors into cutting all ties? Will that extend to someone like google who provides compute to anthropic? Will the administration just plain ignore any court ruling? (as they've shown they're ready to do recently with the tarrifs situation)
If the legal system works as intended, the blast radius isn't too big here and something Anthropic will accept even if it hurts them. Maybe they even win and get the supply chain risk designation lifted. But I have zero faith that the legal system will make a difference here. It all comes down to how far the administration wants to go in imposing it's will.
They can also classify it as restricted data -- like nuclear weapons technology.
Sure, there will be a court battle, but I don't think these companies want to take that chance. They'll capitulate after the lawyers realize that option is on the table.
> They can also classify it as restricted data -- like nuclear weapons technology.
Nuclear weapons technology is restricted under very specific legislative authority, where is the corresponding authority that could be selectively applied to a particular vendors AI models or services?
agreed but the current administration is pretty adept at using the slimmest margin for justification and benefiting from the fact that the legal process playing out over years is extremely detrimental to everyone but the government
> If you are a Department of War contractor, this designation—if formally adopted—would only affect your use of Claude on Department of War contract work. Your use for any other purpose is unaffected.
/In theory./
In practice, if your biggest customer tells you to drop Anthropic, you listen to them.
Remember when A16Z and a bunch of other muppets insisted they had to back Trump because Biden was too hostile to private companies, especially AI ones? Incredible.
> Sam Altman told OpenAI employees at an all-hands meeting on Friday afternoon that a potential agreement is emerging with the U.S. Department of War to use the startup’s AI models and tools, according to a source present at the meeting and a summary of the meeting seen by Fortune. The contract has not yet been signed.
You're absolutely right to point that out -- thank you for catching it. I made a mistake in my previous response and that last act appears to have caused civilian casualties. Let me take a closer look and clarify the correct details for you.
(Will leave you to imagine the bullseye emoji, etc.)
You know what? I have not seen an American company take a stand like this… uh ever. I don’t think there should be any engagement with the military what so ever but I will offer a kudos to Anthropic.
I don’t really expect this to last but if it does I will happily continue to offer this kudos on an indefinite basis.
Could this escalate to the point that Anthropic exits the US and sets up shop elsewhere? Or would the company cease to exist before it got to that point?
It gets so much money, compute and US user data. It won’t be allowed to operate as is as a foreign entity
Best scenario it will get TikTok-ed, otherwise it will become the real national security risk
Had the exit happen, well, as US has a monopoly of compute on this planet for next 2-3 years at least, the company, even though they would take the researchers with them, will certainly cease to exist as it exists now.
Would the US government attempt to apply export controls on the technology and prohibit this? I'm sure Lockheed Martin couldn't decide to move their proprietary technology to another country.
Hegseth's statement already leans towards accusations of treason and duplicity, I would say people trying to export the company would face significant risk of arrest or worse.
This an extremely polite “fuck you, make me”. It’s good to see that they have principles, and I suspect strongly that Anthropic will come out on top here if they stand firm.
If the Trump admin so chooses, they could absolutely obliterate Anthropic in an instant. They don't really care about tricky things like 'legality' or 'the court of law', they could just force everyone to stop interacting with them, raid their offices and steal all their shit.
Perhaps they should've found their spine a year earlier; right now their only hope is that the admin isn't stupid enough to crash the propped-up economy over petty bullshit. But knowing how they behave, well.
> They don't really care about tricky things like 'legality' or 'the court of law', they could just force everyone to stop interacting with them, raid their offices and steal all their shit.
This is criticism that I would use to describe countries like China and Russia, and many other poorer ones. Were the Trump administration to do this, it would be unequivocal evidence that we are dealing with an unlawful insurgent government. I doubt it will happen, but I'm often wrong.
This is all stuff they've already done in the past few months alone. I think it's time for people to take their heads out of the sand and look what's been happening around them.
How long can we push this narrative? It was a terrible situation and I can't imagine the minutes of complete fear she must have felt. I pray for her family. But to then draw a conclusion to say this is evidence that we are in some sort of fascist decline, because of this incident, takes away from the innocent lose of life. And greatly exaggerates the skill and aptitude of the killer. People spew the fascist narrative every chance they get. I'm sure most of us who like strawberries will be picking strawberries come June.
This is what real leadership looks like. Not the silence and complicity that you see from big tech, who regularly bend the knee and bestow bribes and gifts onto the Trump administration.
Title is off: "Statement on the comments from Secretary of War Pete Hegseth"
This is another statement, to their customers about Hegseth's social post, but perhaps resulting in further escalation because you know the other side doesn't like having their weaknesses pointed out.
This applies to basically every military and company in every country in all of human history. Nearly every single other country tries to spy on every single other country, including on the US. That's just how these things go.
This letter is a public part of the negotiation process. It shouldn't be surprising that they are primarily using arguments that are, at least on the face, "patriotic".
Doesn't NSA have a backdoor to all these companies by default? I could have sworn I read somewhere years ago that the government demands a backdoor to all US companies if they can't get in on their own.
1) The US gov generally does have close partnerships with most large-scale, mature tech companies. Sometimes this is just a division dedicated to handling their requests, often it’s a special portal or API they can use to “lawfully” grab information from for their investigations. Often times these function somewhat like backdoors. Anthropic is large, but not mature. Additional changes must still take place for “backdoor” style partnerships to be effected.
2) The NSA can pretty much use any computer system they set their eyes on - famously including computers that were never connected to the internet secured in the middle of a mountain (Stuxnet). If they wanted to secretly utilize the Claude API without Claude finding out, that is within their capabilities. Google had to encrypt all their internal datacenter traffic to try to prevent the NSA from logging all their server-to-server traffic, after mistakenly thinking their internal networks were secure enough not to need that.
3) This isn’t about being “able” to do whatever the administration wants. This is the administration demonstrating the consequences of perceived insubordination to make other companies think twice about ever trying to limit use of corporate technology.
On point 3, are you saying this will dissuade other companies from taking Anthropic's stance? Somehow I actually thought this would set precedent for how to actually stand up to gov. Quite interesting how we see the same situation and come up with totally different conclusions.
A backdoor is a completely different thing when it comes to an AI company, as compared to a social media company. Not really even sure what it would mean when it comes to doing inference on an LLM. Having access to the weights, training data and inference engine?
The model of Claude the DoD is asking for more than likely doesn't even exist in a production ready form. The post-training would have to be completely different for the model the DoD is asking for.
NSA legally isn't allowed to spy on US citizens directly, due to the NSA being a US military organization and the Posse Comitatus act prohibits the US military from being used as a US policing force.
It's one of the hidden and forgotten revelations about the Snowden leaks, where he showed that the NSA had a bunch of filters in their top-secret classified systems to filter out communications from US citizens. Those filters exist because of Posse Comitatus.
How does the filter work? Identity first? As in, do they access the data/activity first and stop when they realize the person is a citizen? Otherwise how do they approach it?
I have worked at a number of software companies that would be "interesting" to get access to, with enough intimate information to know if there was a super-sekret backdoor. If "all US companies" had to comply .. well .. I guess I was really lucky to work for those that somehow fell through the cracks.
I think many people on HN have a cynical reaction to Anthropic's actions due to of their own lived experiences with tech companies. Sometimes, that holds: my part of the company looked like Meta or Stripe, and it's hard not to regress to the mean as you scale. But not every pattern repeats, and the Anthropic of today is still driven by people who will risk losing a seat at the table to make principled decisions.
I do not think this is a calculated ploy that's driven by making money. I think the decision was made because the people making this decision at Anthropic are well-intentioned, driven by values, and motivated by trying to make the transition to powerful AI to go well.
[1]: https://news.ycombinator.com/item?id=47174423
[2]: https://news.ycombinator.com/item?id=47149908
Personally, I wish that the political alignment I favour was as Big Tent as Donald Trump's administration is. I think he can get Zohran Mamdani in the room and say "it's fine; say you think I'm a fascist" and then nonetheless get what he wants. But it just so happens that the other side isn't so. So such is life. We lose and our allies dwindle since anyone who would make an overture to us, we punish for the sin of not having been born a steadfast believer.
Our ideals are "If you weren't born supporting this cause, we will punish you for joining it as if you were an opponent". I don't think that's the path to getting what one wants.
You, too, are driven by money. Yet I’m certain you maintain a set of principles and values. Let’s keep the discussion productive yeah?
Anthropic kept referring to Hegseth as "Secretary of War" and the DoD as "Department of War". Which is horseshit. This whole thing is Anthropic flailing.
Do you just expect Anthropic to totally blow up all bridges to the government? What do you actually want them to do?
Reading your comment history I'm not sure they could do anything to satisfy you.
If the government really wants to, it could try building its "Skynet" on open-source Chinese models.. which would be deeply ironic.
Do you see the problem here. Genuinely don't think we would've won WWII if these people were running things back then.
It’s clear that this is mostly a glorified loyalty test over a practical ask by the administration. Strangely reminiscent of Soviet or Chinese policies where being agreeable to authority was more important than providing value to the state.
If we're going by Occam's razor: it's Friday so Pete probably started drinking ~10:30-11am.
The new thing that I know about leading AI companies that aren't Anthropic (i.e. OpenAI, Google, Grok, etc) is that they knowingly support using their tools for domestic mass surveillance and in fully autonomous weapon systems.
“To the best of our knowledge, these exceptions have not affected a single government mission to date.”
I had assumed these exceptions (on domestic surveillance and autonomous drones) were more than presuppositions.
I've been seeing it a lot lately, but don't remember ever really seeing it before. Do members of the military prefer this title?
We have soldiers, sailors, airman/women, Marines (who really do not like being called soldiers), Coast Guardsman/women, and now the Space Force. Granted, I do not know why "service member" did not catch on. Perhaps because "warfighter" is a bit shorter.
edit: To be clear, Hegseth didn't create it, merely has popularized its use recently. Eg his speech at Quantico last Sept
The term—and its use in the now-Department of War—dates back to the late 80s.
https://en.wiktionary.org/w/index.php?title=warfighter&actio...
The term dates back decades.
It's also incoherent that the DoD/DoW was threatening to invoke the Defense Production Act OR classifying them as "supply chain risk". They're either too uniquely critical to national defense OR they're such a severe liability that they have to be blacklisted for anyone in the DoD apparatus (including the many subcontracts) to use.
How are other tech companies supposed to work with the US government and draw up mutual contracts when those terms are suddenly questioned months later and can be used in such devastating ways against them? Setting the morals/principals aside, how does this make for rational business decision to work with a counterparty that behaves this way.
It's an honest question by the way - not trying to throw any gothas.
Just trying to understand if comoanies or people that don't orbit defense contracting are free to operate with Anthropic still or risk being sanctioned too.
So yeah, extremely few have.
Democracy isn't dead folks, but it takes more work than usual.
[1] https://en.wikipedia.org/wiki/Learning_Resources,_Inc._v._Tr...
Perhaps it’s time or even past time to think of ways of screwing up their training sets.
"Secretary Hegseth has implied this designation would restrict anyone who does business with the military from doing business with Anthropic. The Secretary does not have the statutory authority to back up this statement. 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.
In practice, this means:
If you are an individual customer or hold a commercial contract with Anthropic, your access to Claude—through our API, claude.ai, or any of our products—is completely unaffected. If you are a Department of War contractor, this designation—if formally adopted—would only affect your use of Claude on Department of War contract work. Your use for any other purpose is unaffected."
If the legal system works as intended, the blast radius isn't too big here and something Anthropic will accept even if it hurts them. Maybe they even win and get the supply chain risk designation lifted. But I have zero faith that the legal system will make a difference here. It all comes down to how far the administration wants to go in imposing it's will.
Bleak.
GCP and AWS cannot use Claude to build anything part of a DoD contract, but they do not need to deny Anthropic access to compute itself.
Surely that would cover both buying things from and selling things to Anthropic.
Sure, there will be a court battle, but I don't think these companies want to take that chance. They'll capitulate after the lawyers realize that option is on the table.
Hopefully their lawyers read HN comments so they can negotiate with your deeper understanding of the legal landscape.
Nuclear weapons technology is restricted under very specific legislative authority, where is the corresponding authority that could be selectively applied to a particular vendors AI models or services?
/In theory./
In practice, if your biggest customer tells you to drop Anthropic, you listen to them.
It’s the library of Alexandria all over again.
> Sam Altman told OpenAI employees at an all-hands meeting on Friday afternoon that a potential agreement is emerging with the U.S. Department of War to use the startup’s AI models and tools, according to a source present at the meeting and a summary of the meeting seen by Fortune. The contract has not yet been signed.
https://news.ycombinator.com/item?id=47188698
Fuck this authoritarian bullshit.
You're absolutely right to point that out -- thank you for catching it. I made a mistake in my previous response and that last act appears to have caused civilian casualties. Let me take a closer look and clarify the correct details for you.
(Will leave you to imagine the bullseye emoji, etc.)
I don’t really expect this to last but if it does I will happily continue to offer this kudos on an indefinite basis.
Best scenario it will get TikTok-ed, otherwise it will become the real national security risk
Had the exit happen, well, as US has a monopoly of compute on this planet for next 2-3 years at least, the company, even though they would take the researchers with them, will certainly cease to exist as it exists now.
Hegseth's statement already leans towards accusations of treason and duplicity, I would say people trying to export the company would face significant risk of arrest or worse.
Perhaps they should've found their spine a year earlier; right now their only hope is that the admin isn't stupid enough to crash the propped-up economy over petty bullshit. But knowing how they behave, well.
This is criticism that I would use to describe countries like China and Russia, and many other poorer ones. Were the Trump administration to do this, it would be unequivocal evidence that we are dealing with an unlawful insurgent government. I doubt it will happen, but I'm often wrong.
This is another statement, to their customers about Hegseth's social post, but perhaps resulting in further escalation because you know the other side doesn't like having their weaknesses pointed out.
That’s okay! The use of autonomous weapons is only risky for the civilians of the country you’re destabilizing this week!
1) The US gov generally does have close partnerships with most large-scale, mature tech companies. Sometimes this is just a division dedicated to handling their requests, often it’s a special portal or API they can use to “lawfully” grab information from for their investigations. Often times these function somewhat like backdoors. Anthropic is large, but not mature. Additional changes must still take place for “backdoor” style partnerships to be effected.
2) The NSA can pretty much use any computer system they set their eyes on - famously including computers that were never connected to the internet secured in the middle of a mountain (Stuxnet). If they wanted to secretly utilize the Claude API without Claude finding out, that is within their capabilities. Google had to encrypt all their internal datacenter traffic to try to prevent the NSA from logging all their server-to-server traffic, after mistakenly thinking their internal networks were secure enough not to need that.
3) This isn’t about being “able” to do whatever the administration wants. This is the administration demonstrating the consequences of perceived insubordination to make other companies think twice about ever trying to limit use of corporate technology.
On point 3, are you saying this will dissuade other companies from taking Anthropic's stance? Somehow I actually thought this would set precedent for how to actually stand up to gov. Quite interesting how we see the same situation and come up with totally different conclusions.
The model of Claude the DoD is asking for more than likely doesn't even exist in a production ready form. The post-training would have to be completely different for the model the DoD is asking for.
It's one of the hidden and forgotten revelations about the Snowden leaks, where he showed that the NSA had a bunch of filters in their top-secret classified systems to filter out communications from US citizens. Those filters exist because of Posse Comitatus.