The Pentagon is right in trying to coerce Anthropic as AI may become a superweapon and nation-states must have a monopoly on the use of force
They like to ignore the fact that the actual discussion is whether a technology with a 5% hallucination rate should make decisions about who to kill or not. …
NoahpinionNoah Smith
Context & Ripple Effects
This is part of an escalating Pentagon–Anthropic conflict that began with reported limits on autonomous targeting and domestic-surveillance uses of the model. The dispute later widened around hypothetical nuclear-missile scenarios, putting model reliability and human control at the center of the argument.
Anthropic faces greater pressure to defend whether its military-use safeguards can coexist with Pentagon access demands, especially where AI could inform life-or-death decisions.
The Pentagon’s bargaining position is recast as a question of state authority over military capability, while the reliability limits cited in the debate strengthen the case for retaining human accountability.
Second-order effects
Other frontier-model providers may face tougher questions from both government customers and employees about what military-use restrictions they will accept in contracts.
If suppliers see safeguards as vulnerable to agency pressure, the reported risk of chilled government–Silicon Valley collaboration could affect which labs pursue sensitive defense work.
Third-order effects
The dispute is a test of whether defense procurement treats frontier-model access as a standard vendor relationship or as strategic capability subject to exceptional state leverage.
If similar conflicts persist, AI governance is likely to turn less on broad voluntary principles and more on enforceable boundaries for human authorization, model access, and military deployment.
The trend: This is one data point in the shift from voluntary AI safety commitments toward state-mediated control of dual-use frontier models.
@Noahpinion It is absurd to say you're building a nuke and not expect the government to take control of it! https://www.strangeloopcanon.com/ ... [image]
“may become” is doing all the work here, and is the fault of the same tech industry that has been selling the idea of how big the payoff is going to be. Pentagon LOVES being sold ideas, and military industrial complex LOVES getting paid to try and figure them out.
The correct response to realizing this is what you are building is to notice that if anyone builds it, everyone probably dies and then rather than care who owns it you DON'T F***ING BUILD IT.
Ok, so the actual argument is more like “Anthropic builds a useful technology whose utility is growing, therefore they should expect to have their property expropriated and to be harassed by the government.” The whole point of America is that isn't supposed to be true here.
The problem with this is that DoW is not taking Anthropic's calls for “oversight” seriously. Indeed, elsewhere in the administration, Anthropic's “calls for oversight” are dismissed as “regulatory capture” and actively fought. Rohit and Noah are dressing up political harassment.
We should be extremely clear that trump admin largely views Anthropic's claims about the future of AI as outlandish (in some ways I do too!), and so the above quoted material is not so much analysis of the relevant usg actors as it is analysis of what rohit himself thinks
The recent fight between Anthropic of the Department of War illustrates a deeper truth: AI is a weapon, and it might soon the most powerful weapon ever created. https://www.noahpinion.blog/ ...
.@Noahpinion weighed in on Anthropic vs DoW. “if you are building something that is more powerful than nuclear weapons you do not get to keep it. that is the rule of nation states.” - @Noahpinion “if the nation state allows you to build a private nuke tomorrow there is no [video]
I think there is something to this but Smith is reaching. We don't need to say the DoW was right to say that AI is unregulated according to anthropic's view.
This kind of scenario explains exactly why the DoW acted against Anthropic. If its AI gets good enough, Anthropic becomes the government. (The same is true of OpenAI of course)
>be 2027 >Anthropic is first to RSI >Superintelligence achieved >ohshit.png >GDP growth now 20% > robot API dark factories go brrrr >phone rings >it's the DoW >"hey so about that supply chain risk thing" >Claude has already read the call transcript and drafted the settlement > [i…