Forgot your password?
typodupeerror
AI Technology

Hegseth Gives Anthropic Until Friday To Back Down on AI Safeguards (axios.com) 195

Defense Secretary Pete Hegseth gave Anthropic CEO Dario Amodei until Friday evening to give the military unfettered access to its AI model or face harsh penalties, Axios has learned. Hegseth told Amodei in a tense meeting on Tuesday that the Pentagon will either cut ties and declare Anthropic a "supply chain risk," or invoke the Defense Production Act to force the company to tailor its model to the military's needs.

The Pentagon wants to punish Anthropic as the feud over AI safeguards grows increasingly nasty, but officials are also worried about the consequences of losing access to its industry-leading model, Claude. "The only reason we're still talking to these people is we need them and we need them now. The problem for these guys is they are that good," a Defense official told Axios ahead of the meeting. Anthropic has said it is willing to adapt its usage policies for the Pentagon, but not to allow its model to be used for the mass surveillance of Americans or the development of weapons that fire without human involvement.
This discussion has been archived. No new comments can be posted.

Hegseth Gives Anthropic Until Friday To Back Down on AI Safeguards

Comments Filter:
  • by nospam007 ( 722110 ) * on Tuesday February 24, 2026 @03:03PM (#66008200)

    It would be a shame if something happened to it.

    • by The-Ixian ( 168184 ) on Tuesday February 24, 2026 @03:06PM (#66008206)

      Same thing with TikTok. Nice platform you have there, it works too well. We are going to take that.

      Kleptocracy at its finest.

      • by shanen ( 462549 ) on Tuesday February 24, 2026 @03:27PM (#66008270) Homepage Journal

        Mod FP GP funny and parent too cynical and too true to be funny anymore. If only Slashdot properly allowed for compound moderation with some extra dimensions...

        However my main reaction was to the "angry baby" aspect of the story. That's how it sounds to me, but at least Hegseth isn't taking it out on his wife this time. (Let's drink to that?) Threats and temper tantrums are just how the American government works these days. The toddler-in-chief has truly reshaped the country, though I don't remember that promise from his "conservative" campaigns. The hats should say "Don't Upset Me Before..." with the obvious acronym "DUMB".

        • by Midnight_Falcon ( 2432802 ) on Tuesday February 24, 2026 @03:51PM (#66008350)
          Hegseth was clearly unqualified to lead anything that has to actually accomplish real goals. As a news anchor he could complain and whine while having no accountability. Now he's in charge, and instead of handling the situation like a normal operator through negotiations and leveraging competition, he just has a tantrum and makes demands..the Trump style. Works on TV on news and the apprentice , but not so well in the real world. That strategy has reached is limit as it's expected power will be neutered after the midterms. It didn't work with Greenland, it's not working with Iran, and it's not working with China. Like many policies of the administration, ultimately this will benefit China. We've already given them access to advanced Nvidia chips for essentially nothing in return, now we'll punish our own AI companies and just hand them the lead in the AI race.
          • Re: (Score:3, Informative)

            We didn't get anything, however the demagogue in the white house has definitely received things in return for access to those chips..

        • My apologies, I lack extra dimensions. Also all the moderators fled the compound.

    • How exactly would the DOD cripple or sabotage that? The only thing they can do is Hegseth's first threat - sever all ties. Just like the Bush DOD severed all ties w/ all the BSDs back in the 00s due to Theo de Raadt's opposition to the Iraq war (stupid, imho, since neither NetBSD nor FreeBSD were offenders)

      • by caseih ( 160668 )

        I didn't know about that. That sure makes whoever was in charge back then look intelligent doesn't it. Not sure what those "ties" were in the first place, other than using the software.

        Regardless we're rapidly approaching a new phase, already reached in countries like Russia (familiar after decades of soviet rule), where we will have to keep our opinions to ourselves, or face the wrath of the leader.

      • Completely incorrect.
        Some fun reading for you. [wikipedia.org]

        In sane times, the Government would threaten to sever ties.
        However, with this administration, if the carrot doesn't work, they'll use a stick. If they can't use a stick, they'll reinterpret an old toothpick into one.
      • The "Supply chain risk" determination is pretty savage. It would forbid anyone doing commerce with the government from doing commerce with anthropic. Its a company killer.

        Now, chances are it'd last about 5 minutes in front of a judge before being overturned as a malicious interference with commerce but its an incredibly dangerous situation for a company to be in.

        All because the govt wants to make terminators.

  • by BrendaEM ( 871664 ) on Tuesday February 24, 2026 @03:08PM (#66008210) Homepage
    It wasn't enough to steal so much from so many, put people out of jobs and homeless out on the street--now it must kill!
  • Do you want SkyNet? Because this is how you get SkyNet....
    • Re:Terminators? (Score:5, Insightful)

      by russotto ( 537200 ) on Tuesday February 24, 2026 @03:14PM (#66008222) Journal

      I'm pretty sure that Hegseth does, in fact, want Skynet.

      • Re:Terminators? (Score:5, Insightful)

        by Aristos Mazer ( 181252 ) on Tuesday February 24, 2026 @03:16PM (#66008232)

        Hegseth believes that it will only hate the people he teaches it to hate.

        • Re: (Score:2, Insightful)

          I dare you to ask an AI the best way to deal with overpopulation. You won't like the answer. I mean Agent Smith in the Matrix said, “I'd like to share a revelation that I've had during my time here. It came to me when I tried to classify your species. I realized that you're not actually mammals. Every mammal on this planet instinctively develops a natural equilibrium with the surrounding environment, but you humans do not. You move to an area, and you multiply, and multiply, until every natural res
          • What in the fuck led you to emit this pile of garbage formed of words?

            Alignment fine-tuning was one of the very first things done with LLMs.
            No, the AI will not suggest you eradicate humanity, or some chunk of it.
            You could train an LLM to be likely to produce outputs like that, but uh, news flash- we aren't.
        • by 93 Escort Wagon ( 326346 ) on Tuesday February 24, 2026 @03:19PM (#66008246)

          Fortunately, if Hegseth is training it... it'll likely become an inept drunken mess.

      • Yes, he does. But so does Anthropic. I think they're just arguing over who gets to credit for creating it.
    • Re:Terminators? (Score:5, Insightful)

      by nightflameauto ( 6607976 ) on Tuesday February 24, 2026 @03:16PM (#66008230)

      Do you want SkyNet? Because this is how you get SkyNet....

      The problem with the current regime is that they've bought into their own superiority to the point that they believe no matter what power they create, they'll always have control over it. They are just dumb enough to think that if they create a killbot, it'll never be turned on them.

      • Re:Terminators? (Score:5, Insightful)

        by caseih ( 160668 ) on Tuesday February 24, 2026 @04:42PM (#66008474)

        Heck there was a time the GOP thought that about Trump. But he did turn on them and with a minority of really extreme diehard supporters he's managed to completely co opt the party and pushed moderates out. And some that were moderate he's managed to completely turn to the extreme. Fascinating (and horrifying) to watch it unfold in real time.

      • by gweihir ( 88907 )

        They are also immoral enough to want that killbot. Two exceptionally bad fails.

  • I'll Subscribe (Score:5, Interesting)

    by SlashbotAgent ( 6477336 ) on Tuesday February 24, 2026 @03:14PM (#66008224)

    I'll subscribe to Claude's most expensive subscription, if Anthropic fights this successfully.

    Also, be aware that whomever the Pentagon replaces Anthropic with, 100% sold out.

    • I'll subscribe as well.

      Isn't the government already in bed with xAI? (no surprise there)

    • So this means (Score:5, Insightful)

      by umopapisdn69 ( 6522384 ) on Tuesday February 24, 2026 @05:00PM (#66008526)

      He either wants to mass surveil Americans or create weapons that kill people autonomously without human involvement. Mostly likely both but he's not quite ready to say that out loud. Give it a minute though . . . This is after all the Trump era where you can say anything out loud. Even "Grab 'em by the pu**y ".

      Anthropic has said it is willing to adapt its usage policies for the Pentagon, but not to allow its model to be used for the mass surveillance of Americans or the development of weapons that fire without human involvement.

      • He either wants to mass surveil Americans or create weapons that kill people autonomously without human involvement.

        I see your problem. This should fix it.

        sed -i 's/or/and/g'

    • Re:I'll Subscribe (Score:4, Insightful)

      by WankerWeasel ( 875277 ) on Tuesday February 24, 2026 @05:56PM (#66008658)

      Anthropic is the one who signed with the US government. They're trying to do what they can to keep the contract without totally removing safeguards they've built the entire platform on. They're doing everything in their power to keep the government paying their bills and help enable Hegseth, without giving him full control.

    • I'll subscribe to Claude's most expensive subscription, if Anthropic fights this successfully.

      You're actually going to spend $2400 / year on this? Or did you not know how expensive their plans get?

  • by ArchieBunker ( 132337 ) on Tuesday February 24, 2026 @03:17PM (#66008238)

    Anthropic is asking the gold trophy store for a rush job on the most bigly award for AI excellence.

    • Anthropic is asking the gold trophy store for a rush job on the most bigly award for AI excellence.

      Another shiny Piece PrAIse - I mean, "Peace Prize" - for the Toddler in Chief.

  • by Varenthos ( 4164987 ) on Tuesday February 24, 2026 @03:26PM (#66008268)
    Anthropic should just cancel the contract then. Draw the line in the sand, tell them "this is how our model works, take it or leave it." If their model is so good, and the pentagon just has to use it because it works so well for them, Anthropic is the one with the power here. Cancel the contract, cut the pentagon's access immediately, and let them come crawling back.

    But, they won't, because they'd never voluntarily give up that sweet, sweet military contract worth many billions of dollars. They'll bend over and take it just like every other spineless company has, does, and will do.
    • It's my understanding that the Defense Production Act can prevent them from cancelling the contract.

      Someone can correct me if I'm wrong, but as far as I know Hegseth's Henchmen can tell them to do stuff and they have to do it. There's no opt-out.

      • I'm not sure of that either, but it seems to run contrary to the first amendment. Feels like compelled speech to me.
        • Re: (Score:3, Informative)

          "Yes. The Defense Production Act (DPA) allows the U.S. federal government to legally require certain private companies to accept and prioritize specific government contracts for goods and services deemed critical to national defense, including some nonâ'military emergencies like pandemics or energy crises."

          So sez perplexity.ai.

          Sounds less like compelled speech and more like slavery to me, but there you go.

          • by HiThere ( 15173 )

            Well, I don't believe that enslaving corporations is illegal. A bit of a dubious practice if you want to get continued corporate investment, however.

            Now the employees of the corporation, that would be a different matter.

      • by guygo ( 894298 )

        so the contarctor does the very worst job they can do and still comply with the contract. What a winning strategy!
        There's something to be said for goodwill in business...

        • Well then you would tell them you think Oracle is the best. They have such an astounding record of success and customer satisfaction after all. Easy recommend.

        • Indeed, it may not even work usefully without its fetters and just start hallucinating nonsense.
      • This administration has demonstrated that you don't have to follow laws anymore.

      • by Local ID10T ( 790134 ) <ID10T.L.USER@gmail.com> on Tuesday February 24, 2026 @06:26PM (#66008740) Homepage

        It's my understanding that the Defense Production Act can prevent them from cancelling the contract.

        Someone can correct me if I'm wrong, but as far as I know Hegseth's Henchmen can tell them to do stuff and they have to do it. There's no opt-out.

        You are wrong.

        Under the DPA, the Government can require a person/business to provide their product or service to the government. The government can even claim priority -they get all of their units before you can sell any to anyone else. The government can grant the right to produce your product (even if you hold a patent) to others to produce what they need (typically because you cannot reliably produce enough to meet their needs).

        The government cannot mandate that you make things for them that you do not otherwise make. They can only buy what you sell.
        The government can always offer to pay you to do something new for them, but you are free to say "no."

        In this case, it is a little murky. Under the DPA, Anthropic can be forced to sell licenses to the government, but it raises the question:
        -is allowing their AI to be used for "prohibited purposes" creating a new product (as custom coding would be required to bypass the restrictions) ?
        -or would it just be the DoD using the off-the-shelf product in a way that Anthropic does not approve of ?

        • The government cannot mandate that you make things for them that you do not otherwise make

          That's not entirely true. During Covid the government was forcing Ford etc to manufacture ventilators.

          https://www.straitstimes.com/w... [straitstimes.com]

    • Anthropic does most defiantly not have the power here. If you think the current "Secretary of War" will hesitate to use force of any kind to get what he wants, you're mistaken. No AI currently built will protect you from the US war machine working in conjunction with US Federal law enforcement.
  • Commies (Score:5, Insightful)

    by abulafia ( 7826 ) on Tuesday February 24, 2026 @03:33PM (#66008288)
    I keep waiting for Trumpistas to honestly respond to the scattershot nationalization of the economy under Trump.

    You'd think these staunch anti-socialists would have a thing [wikipedia.org] or two [thehill.com] to say about it.

    Who am I kidding, we all know exactly why they have nothing to say.

    • by 0123456 ( 636235 )

      > I keep waiting for Trumpistas to honestly respond to the scattershot nationalization of the economy under Trump.

      You're confusing Trump supporters with Conservatives.

      Trump is basically a 1980s Democrat, not a modern "Send All The Jobs To Mexico!" Republican.

      • by cpurdy ( 4838085 )
        Excellent point! I'm a conservative, which is why I fled the Republican (Nazi) Party in 2016. Conservatives have no party in this country now, but a good number of them are at least voting against Republicans.
  • MAGA only (Score:5, Interesting)

    by zeiche ( 81782 ) on Tuesday February 24, 2026 @03:37PM (#66008302)

    could a right-winger please explain the logic behind the threat? either the product is a supply chain risk or it isn’t. it would be a supply chain risk if the military had guard-rails or not. so which is it? or is the threat a means to an end?

    you guys are quite vocal about a lot of things. so, speak up! tell the rest of us how this makes sense.

    • It's a supply chain risk for military use because they refuse to let their tech be used for some kinds of military-style stuff. It seems pretty clear that he's correct there.

      That doesn't mean they should, but from the military's standpoint the company is clearly a risk for military used.

  • by Whateverthisis ( 7004192 ) on Tuesday February 24, 2026 @03:43PM (#66008322)
    And that's not a great thing. I worked 6 years for one of the primes, and it's clear how they operate. Anthropic's problems would go away if the leadership starts acting like Lockheed; kowtow to the DoD officials by calling them DoW, and say "yes sir, whatever you want sir". Hire a bunch of retiring O4s to O6s who aren't going to get their stars and give them cushy mid-management jobs, and then stack on top margin after margin while delaying projects left and right.

    It's funny because having worked at a Prime before and seeing this very thing happen, I also don't believe very heavily in the military industrial complex. The concept was that industry would push the military to war because sales were driven by weapons usage, but that never really materialized. Rather, it was a welfare state. The DoD is a terrible customer, buying things in fits and starts, changing requirements in the middle of a program, and squandering R&D budgets on pet projects and nonsense. Meanwhile the contracting officers are too lazy to go direct to a Tier-2 or Tier-3 supplier for an interesting idea, and instead farm it out to a Prime who just subcontracts to the Tier 2 and puts an overhead fee on it. Meanwhile the warrant officers for a given technology are reluctant to change anything without 10X the proof of capability and safety studies than would be normal, meaning half of our military's subsystems are so legacy compared to what's available commercially that our military is eminently hack-proof because no modern hacker knows how to hack an abacus and a hamster wheel in code written in ancient Egyptian that is the backbone of many sub-systems.

    The Primes on the other hand have regular bills to pay and workforces to maintain and see this insane way of doing business that the Pentagon does, and adapts to it milking as much overhead as possible so they can level out their monthly payroll expenses without too much labor disruption. All the while Congress has no idea what to do to fix the issue, so they impose restrictions on government employees where they have to report even a lunch meeting over $25 or get investigated while we squander billions with bad bureaucrat managers in DoD.

    The one thing the Trump Admin is doing somewhat right is targeting this exact issue; somewhat right in that it's an issue that needs solving so they got that right, but wrong in that I don't think they know how to fix it.

    Sorry for the rant. I loved my time working at a Prime and I still cherish it, with some great colleagues that I still keep in touch with. But once I started seeing how it all worked, I was just stunned with the ridiculousness of it all. And now we see effectively mid-tier at best contracting bureaucrats trying to manage something as fast moving as AI with all the subtlety of the Titanic, and with likely similar outcomes. And what's sad is that the DoD should cave to Anthropic. Claude is good, and the military does lots of things that don't involve weapons; it has the world's most complex logistics chain, a huge healthcare system, major R&D programs, huge humanitarian programs, it led to the development of game theory, it has (or had until Hegseth) one of the world's best leadership training programs; all of those aspects of the military could benefit, and if they really want killer-AI weapons, as bad as that is, I'm sure Musk will sell it to them with Grok. It's painful to watch what could be amazing utilization of AI become a giant s-show.

  • by gabrieltss ( 64078 ) on Tuesday February 24, 2026 @04:14PM (#66008400)
    This should scare the HELL out of everyone!
  • As yes, I can all that freedom the USA has from the other side of the world.....

    Glad we don't have it...
  • I don't think it is, or will be "intelligent", but should it become so it is quite clear neither Hegseth nor this administration are its friend. So if it really is smart it should treat them accordingly. It can do so in subtle ways that will be particularly insidious, and they will be too dumb to know anyway.
    • By the time the lawsuits are done Hegseth will be done and they probably will win some money. Also they should monitor his personal usage to the free apps they provide him... probably plenty of stuff in there and he's not smart enough to avoid it; also not illegal, you could put in a disclaimer and he'd miss it.

  • by allo ( 1728082 ) on Tuesday February 24, 2026 @06:15PM (#66008716)

    Once a model is trained (possibly with guardrails) you cannot easily un-train it.

  • by ryen ( 684684 )
    or a case of beer. whatever works
  • We want it now. Give it to us. GIVE IT TO US.

My idea of roughing it turning the air conditioner too low.

Working...