Anthropic: Stay strong!
I don’t have time to write a full post right now, but hopefully this is self-explanatory.
Regardless of their broader views on the AI industry, the eventual risks from AI, or American politics, right every person of conscience needs to stand behind Anthropic, as they stand up for their right to [checks notes] not be effectively nationalized by the Trump administration and forced to build murderbots and to help surveil American citizens. No, I wouldn’t have believed this either in a science-fiction movie, but it’s now just the straightforward reality of our world, years ahead of schedule. In particular, I call on all other AI companies, in the strongest possible terms, to do the right thing and stand behind Anthropic, in this make-or-break moment for the AI industry and the entire world.
Follow
Comment #1 February 27th, 2026 at 3:47 pm
Look, let’s not pretend this is something it isn’t.
The government isn’t demanding that AI companies build “murderbots” for the sake of it. This is about national defense—plain and simple. We’re in a strategic competition with China and Russia, and pretending otherwise doesn’t make the threat disappear.
We’ve seen this before. During the Cold War, there were plenty of voices arguing that the U.S. should abstain from nuclear weapons on moral grounds. If we had, we wouldn’t be patting ourselves on the back—we’d be living under a very different world order. Deterrence worked because both sides understood the consequences. That’s what mutually assured destruction is.
Now we’re entering a new phase. Superintelligent AI has the potential to shift that balance just as dramatically. Faster response times, cyber capabilities that could disable enemy systems, autonomous defense networks, drone swarms—these aren’t hypotheticals anymore. They’re the next generation of strategic deterrence.
If the U.S. steps back, China won’t. That’s the reality. And if they get there first, they won’t hesitate to use that advantage. So the question isn’t whether these systems will exist—it’s who builds them, and under what values.
You can call it uncomfortable. You can call it dystopian. But this is the world we’re moving into. Either the U.S. leads in developing these systems to defend its interests and freedoms, or it falls behind and lets the communists take over. If you believe in America and defending our freedoms and our nation, then you should believe in using AI to protect ourselves national security.
Comment #2 February 27th, 2026 at 4:01 pm
Mike Granger #1: The Pentagon should have simply signed a contract with a different provider then. For the government to strongarm Anthropic into doing its will is authoritarian, and actually worse than anything China has done in AI. If you can’t see that this administration should not be trusted with unlimited power over AI, then you are morally deficient.
Comment #3 February 27th, 2026 at 4:35 pm
My prerequisite for people participating in this conversation is for them to read and understand Zvi. In particular:
(1) Realize that, by saying both that Claude is so essential to the supply chain that Anthropic should be compelled to modify its contract under the DPA and also that it’s a supply chain risk, the Pentagon showed that it has no coherent position here at all, besides wanting to punish and destroy Anthropic for (unlike the administration) having any principles at all.
(2) Realize that the Pentagon has grievously harmed US national security, because after seeing what happened to Anthropic (which had tried to help), no other leading AI company is likely to want to work with the DoD going forward.
Comment #4 February 27th, 2026 at 5:52 pm
EVERY news story on this subject should commence by pointing out that Anthropic and the Pentagon have had a preexisting contract in place since last July that *expressly* includes the two restrictions to which Hegseth is objecting. Anthropic did not just show up on Monday and make a lot of unacceptably woke demands; Hegseth is using threats to try to get them to alter a contract that HE (or one of his minions) already signed.
Comment #5 February 27th, 2026 at 6:00 pm
Naturally, any sane person agrees with you. Just as naturally, the first reply comes from yet another blissful mind confidently mistaking partisan bullshit for facts. One can only hope the midterms arrive before they manage to sink the U.S. for good.
That said, #2 is going a bit too far: China already has mass surveillance, and I wouldn’t be surprised if Jack Ma had happily swapped his government’s pressure for what Dario Amodei is dealing with now.
Comment #6 February 27th, 2026 at 8:01 pm
The actual harm to credibility in national security is sufficient that other AI related security companies such as Palantir really should have crashed.
Comment #7 February 27th, 2026 at 8:29 pm
Scott,
Of course everybody reasonable agrees with you. There is nothing to discuss, in fact the two “red lines” are one technically infeasible, the other clearly unconstitutional, so they are basically non-existent unless whoever is asking (cough cough) is either or both completely incompetent or somebody who uses the constitution as toilet paper.
You should maintain your sanity and close comments which would only attract trolls.
PS: the Zvi piece you linked says that OpenAI has sided with Anthropic, xAI with the gov’t and Google hasn’t taken position yet.
Comment #8 February 28th, 2026 at 12:40 am
Just cancelled my OpenAI subscription. A drop in the ocean maybe, but I don’t want to support a service provider whose capability for mass surveillance is likely most imminently used to subvert your midterm elections. Cambridge Analytica was just a humble start on what all is feasible.
I don’t think every AI provider could be deterred from providing mass surveillance tools to your government – this is the well known coordination problem – but one must do what little one can.
Comment #9 February 28th, 2026 at 5:45 am
I don’t have a good opinion of either trump or his administration.
However, it should “We the people” (and their elected representatives) to decide how AI should be used by government, not “We the CEOs of AI companies”.
Comment #10 February 28th, 2026 at 10:53 am
GD #5
“Just as naturally, the first reply comes from yet another blissful mind confidently mistaking partisan bullshit for facts.”
Excuse me but I didn’t find the tone of the first post in any way conveyed the author was “blissful”. How did you determine he has a blissful mind or is that just another ad hominem excess? Also can you identify what in particular was partisan bullshit?
Mike Granger #1
I believe the issue is more imminent in that foes of the US know the US is relatively casualty averse and so just need to be patient and they will withdraw. If you accept that the US is relatively casualty averse then replacing at risk soldiers with electromechanical AI systems makes eminent sense. In Iran we have witnessed effectively undiscerning biological murder bots patrolling the streets but no appetite in the US for ground casualties to stop it.
In a world with no competition then little risk for the strongest country but the US has competitors with no regard for the philosophy that underlies Western Civilization. As you note best to recognize that in order to develop effective strategies rather than to “blissfully” ignore. Also the case at this time that some just like the idea of the Chinese system in preference to the US,
Comment #11 February 28th, 2026 at 11:51 am
Of course I agree with all the comments expressing concern about mass surveillance and don’t have enough information to answer why they didn’t just turn to a Claude alternative. I had the same question when I heard about this.
If you accept the premise that AI will change the world somewhat less, as much, or more than atomic weapons then certainly the government has a legitimate interest. The calls previously have been that the government provide comprehensive legislation for AI and potentially that could include using any AI for national security.
Comment #12 February 28th, 2026 at 12:15 pm
Scott,
“no other leading AI company is likely to want to work with the DoD going forward.”
Sam Altman says, “Hold my beer!”
https://x.com/sama/status/2027578652477821175?s=20
Comment #13 February 28th, 2026 at 1:59 pm
Manta #9: It is we the people who decide (or at least, who vote for those who do). But we also have freedom of contract in the US, and that was the issue here!
Comment #14 February 28th, 2026 at 6:25 pm
Khameini is transiting Styx about now.
Scott #13
Right to freely contract is effectively abrogated with a reasonable finding of in the national defense interest I believe. Advanced software, hardware, strategic minerals, etc have additional controls and the federal government can establish its own priorities. Of course no experience as to an AI, and sure Anthropic would have great legal representation, but just wanted to point out that if sufficiently important for national security then the right to freely contract is significantly reduced.
Comment #15 February 28th, 2026 at 7:13 pm
This appears to be yet another Anthropic publicity stunt, where the company tries to convince everyone that it, and it alone, has the wisdom and expertise to deal with the coming AI apocalypse. Anthropic could have just quietly negotiated, as OpenAI did.
Comment #16 February 28th, 2026 at 8:02 pm
Roger Schlafly #15: You have basic facts wrong. Anthropic tried to quietly negotiate. The Pentagon is the one that went to the press.
Comment #17 February 28th, 2026 at 9:35 pm
AT#12
This is ridiculous.
Altman first say the same things as Amodei then (if the news reports are correct) signs the same contract as the now banned company!
Comment #18 February 28th, 2026 at 10:53 pm
Del #17,
“Altman first say the same things as Amodei then“
It’s called lying. Altman will maintain everywhere how deeply he cares about not allowing mass surveillance or autonomous weapons. Everywhere but the actual contract. His talk is very very cheap.
Comment #19 March 1st, 2026 at 2:56 am
Amen!
I’m now trying to get my head around what Sam Altman is doing with https://openai.com/index/our-agreement-with-the-department-of-war/
William Ehlhardt and I were just discussing it the comments of my own post about this: https://agifriday.substack.com/p/anthropolitics/comment/221309278
Excerpts:
> Contrary to OpenAI’s spin, their new contract with the DoW does not have the “same red lines” as Anthropic wanted. If you read the relevant contract sections in OpenAI’s own post […] you will find it is simply “all lawful use” but with more words.
That guy sure is inconsistently candid [this is an inside joke about the 2023 OpenAI board drama, for those just tuning in].
So the allegation is this: Altman is doing this ultra-slippery thing where he says “we support those same red lines” and then in the contract it’s like “specifically, we support those red lines as defined by current law” — and the entire point of Anthropic’s stand was that current law is lagging behind what’s now possible with AI. Ie, it’s currently possible to adhere to the letter while violating the spirit — like by gathering up public data or purchasable data and then mass correlating everything to get dossiers on everyone.
(I’m not sure if there’s a similar argument regarding the other red line, killbots.)
Comment #20 March 1st, 2026 at 10:56 am
Even in Democracies, governments retain sizeable power when it comes to national security. How else can you have export controls on chips (isn’t that strong arming Nvidia), or chip making equipment in Netherlands. Of course Anthropic has the right to walk away from the contract , which they did, and government did find an alternate provider. Of course, if the government deems fit , for national security, it can procure all the Nvidia chips that are produced and create a Manhattan project equivalent to build advanced AI. Also note that government was always saying lawful purposes, and mass surveillance was not part of it. Once a government is elected, choosing to decide its morality leads to nowhere, the only recourse is the justice system
Comment #21 March 1st, 2026 at 12:32 pm
Iran operations ‘ahead of schedule’, says Trump.
I guess that means – the nuclear holocaust is closer than ever.
Pretty!
Comment #22 March 1st, 2026 at 4:15 pm
Reading OpenAI employees these days, I’m having a genuinely hard time telling whether they’re putting on like three layers of irony or whether this is actually what they think.
Boaz Barak (notable, but not the worst such case):
“I’m proud to work at a company that contains people as brilliant and conscientious as Leo and allows them to speak their mind. OpenAI has a lot of issues, but in terms of enabling employee pushback and discussion it is in fact still open, with all the messiness that this entails.
Leo is an amazing researcher and person but not a lawyer or a natsec expert.
For an amazing and conscientious natsec expert who understands these issues at a deeper level, follow @natseckatrina [who is the natsec policy person at OpenAI]”
Comment #23 March 1st, 2026 at 9:04 pm
In related news, Our Friend Mahmoud has been vaporized as a nice holiday gift.
Comment #24 March 2nd, 2026 at 3:28 am
Anthropic is not made of heroes of the free world. They are a company which (probably) correctly assessed (economic) pros and cons in agreeing with the Pentagon’s requests. Given the precedents like Apple and Tariffs, they are 1) confident to block everything in court and 2) loose too much value in the eyes of their other customers in agreeing with the current US administration. They are playing their cards against the DoW. Everybody saw recently that the administration is not invincible. At least for now..
Comment #25 March 2nd, 2026 at 6:01 am
LK2 #24: It’s precisely because Anthropic is made of fallible humans rather than perfect “heroes of the free world,” that we should support them when they do the right thing, in hopes of encouraging them to continue doing the right thing in the future.
Comment #26 March 2nd, 2026 at 6:43 am
Scott #13
If I understand correctly, the government is saying “either do what we want, or we cancel all your government contracts”; correct me if I’m wrong. Which falls squarely under “freedom of contracts”.
That said, I think that framework is misleading for two reasons 1) it’s corporations, not people: so you are saying that contract laws between companies is a moral issue, which I find bizarre. And 2) the other side is government, where the “freedom” framework makes even less sense
Comment #27 March 2nd, 2026 at 8:30 am
Many US weapons systems depend on a communication link with an operator and/or GPS. In recent combat these requirements have rendered weapons ineffective due to effective signal jamming. Drones that act autonomously in a jammed environment are already in use and one apparently gets by with Raspberry Pi.
Questions about autonomous weapons operation frame differently if you are actually at risk on a battlefield vs in a corporate office on the West Coach.
This isn’t cutting edge but already in effective use.
https://www.nytimes.com/2025/12/31/magazine/ukraine-ai-drones-war-russia.html
Comment #28 March 2nd, 2026 at 12:08 pm
Manta #26: Given its unique coercive and taxing power, the government needs to satisfy a higher standard than you or I would, which at the very least includes not acting vindictively. It would be fine for it to cancel this specific contract with Anthropic. But trying to destroy Anthropic by refusing to work with any company that does business with Anthropic, on pretextual grounds where the real grounds are clearly just that Dario refused to bend the knee to Pete Hegseth? That’s self-evidently the behavior of an authoritarian regime.
Comment #29 March 2nd, 2026 at 1:26 pm
Just want to chime in and say this: given how easily the us and israeli militairies have ended Iran’s nuclear program and decapitated the regime, its pretty clear the Iran nuclear deal was a huge mistake.
Despite the batshit insane stuff from Trump and hegseth I’m glad they have power. No one else was willing to pull the trigger.
Comment #30 March 2nd, 2026 at 5:12 pm
Scott # 28
“That’s self-evidently the behavior of an authoritarian regime.”
It seems to me that authoritarian regimes are characterized by more direct and far more persuasive tactics than this. A couple unexpected deaths or disappearances are usually sufficient to garner wholehearted cooperation.
The recent elections certainly show none of the essential hallmarks of authoritarianism.
Truman actually nationalized the US steel industry in 1952 just prior to a United Steelworkers strike. The Supreme Court made him give back the steel mills to their owners.
Comment #31 March 2nd, 2026 at 8:42 pm
meanwhile OpenAI went ahead and signed a contract with the US government to replace Anthropic and without binding the US government to any such policy.
I assume OpenAI has turned onto a hell of a place to work at
The key turning point for the company was when the employees wanted to get back Sam Altman after he was dismissed as CEO cause it was beneficial to their pockets, making them millionaires and more.
These things are like this, once you cave in for the money and ignore the problems with the CEO being a lier, you have set the path you are on.
Comment #32 March 3rd, 2026 at 10:34 am
Google employees petitioning to reject same kind of deal that OpenAI – to its shame – just engaged with hegseth: https://www.cnbc.com/2026/03/03/anthropic-fallout-iran-war-tech-military-ai.html
Comment #33 March 3rd, 2026 at 5:19 pm
#1 Mike Granger
Omg, your comment is literally written by an LLM.
Comment #34 March 4th, 2026 at 5:11 am
I liked the comments at the end of the NYT article. Eric Schmidt noting his hope that autonomous weapons would make invasions too expensive to contemplate thus reducing international conflicts. The journalist noting that John Gatling had the same hope when he manufactured the rapid fire Gatling Gun.
My view is that if you consider Western Civilization worth defending then it requires further development of autonomous weapons. I sometimes suspect that many are comfortable living in Western Civilization but believe defense of it should be left to others that attend to unpleasant details. Worst case is the seemingly increasing number that don’t accept even that Western Civilization should be defended.
Comment #35 March 4th, 2026 at 11:49 pm
#34
arguably our system survives because governments and elite need the rest of us for economic and military reasons.
once they do not, I wonder where we will end up.
if you have mindless robots following orders, they don’t need us for economy and they also don’t need us for wars. therefore the political and economic and military elite will become independent from the rest of population. because they didn’t need us, our influence would be diminished.
Comment #36 March 6th, 2026 at 7:01 am
Anon #35
“ arguably”
I have no basis to argue. 🙂
There are so many variables and so a multitude of possible scenarios. If sufficiently intelligent AI then the great equalizer because then the elites also aren’t really necessary and all the bureaucracies replaced by far more responsive and efficient AI.
I think it is necessary to consider different periods of a general process. First would be say the gradual replacement of workers and soldiers and government bureaucrats. The displaced humans would eventually reach critical mass and would effectively control the government assuming democracy. They would vote their own interests.
Another possibility might be eliminating the excess population by some means. This doesn’t seem likely to me because the numbers would still vastly favor the displaced.
If the top AI doesn’t control its own programming then the source of human power would be who controls its programming. Probably best to have a national vote for program changes. 🙂
Comment #37 March 10th, 2026 at 7:20 am
Amodei recently noted-
“ Engineers have observed activity patterns associated with concepts such as anxiety appearing in specific contexts.”
I wonder if this is related to spiders. My daughters have anxiety about spiders.
Comment #38 March 10th, 2026 at 8:34 am
The technology described in ‘1984’ was apparently provided by OpenAI. Who knew that Big Brother was just the avatar of a corrupt and slippery salesman named Sam Altman?
Comment #39 March 10th, 2026 at 10:57 am
Those who would welcome a surveillance state and being permanently monitored by AI should consider the following. While the data processing will be done by AI, any alerts this may raise will be interpreted by the Police or Homeland Security, which is not AI but BI. What the hell is BI? you may ask. BI stands for ‘barely intelligent’. You have been warned, when you are shot in the back by ICE agents after AI flagged you as a highly dangerous terrorist, because the surveillance camera had a dirty lens or a faulty sensor.
Comment #40 March 10th, 2026 at 3:47 pm
[…] Anthropic’s position—as expressed for example by Noah Smith, or in the comments of my previous post on this. The argument goes roughly like […]
Comment #41 March 10th, 2026 at 4:45 pm
JanSteen #38 #39
Maybe Claude would provide an incremental improvement but the national surveillance ship sailed years ago.
The type of data collected and analyzed at the Bluffdale NSA facility that came online in 2013 has been debated by various whistle blowers and NSA directors. The Snowden documents support the whistleblower accounts.
A few years ago I re-entered the US when there were pilot facial recognition programs running at select airports. When I reached the passport control desk the agent called me by my first name while sporting a grin. Facial recognition is running at all major US international airports now.
It doesn’t seem to matter, the surveillance state grows whichever party is in control.
If Anthropic doesn’t want their AI used for these purposes then understandable but that will have no impact on the probability of me being shot in the back by a government agent.
Comment #42 May 28th, 2026 at 10:01 pm
I guess it would be a fitting outcome if it turns out mathematicians spent hundreds of billions of dollars creating a humanity-crushing machine and then are among the first to be crushed by the machine. Hubris, atē!