Anthropic’s Statement on the Department of War’s Demands
Pete Hegseth, the US Secretary of War, earlier this week ordered Anthropic, the company that makes Claude artificial intelligence products, to allow the Department of War unrestricted use of Claude. Anthropic is resisting, saying that certain restrictions on Claude’s use are necessary to protect “democratic values”.
[Sculpture by Louise Bourgeois]
Certainly some readers will dismiss what follows as a marketing tactic or corporate propaganda, which it may be. But nonetheless, its content raises important and interesting issues.
Here’s the excerpt:
Anthropic understands that the Department of War, not private companies, makes military decisions. We have never raised objections to particular military operations nor attempted to limit use of our technology in an ad hoc manner.
However, in a narrow set of cases, we believe AI can undermine, rather than defend, democratic values. Some uses are also simply outside the bounds of what today’s technology can safely and reliably do. Two such use cases have never been included in our contracts with the Department of War, and we believe they should not be included now:
-
- Mass domestic surveillance. We support the use of AI for lawful foreign intelligence and counterintelligence missions. But using these systems for mass domestic surveillance is incompatible with democratic values. AI-driven mass surveillance presents serious, novel risks to our fundamental liberties. 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. For example, under current law, the government can purchase detailed records of Americans’ movements, web browsing, and associations from public sources without obtaining a warrant, a practice the Intelligence Community has acknowledged raises privacy concerns and that has generated bipartisan opposition in Congress. Powerful AI makes it possible to assemble this scattered, individually innocuous data into a comprehensive picture of any person’s life—automatically and at massive scale.
- Fully autonomous weapons. Partially autonomous weapons, like those used today in Ukraine, are vital to the defense of democracy. Even fully autonomous weapons (those that take humans out of the loop entirely and automate selecting and engaging targets) may prove critical for our national defense. But today, frontier AI systems are simply not reliable enough to power fully autonomous weapons. We will not knowingly provide a product that puts America’s warfighters and civilians at risk. We have offered to work directly with the Department of War on R&D to improve the reliability of these systems, but they have not accepted this offer. In addition, without proper oversight, fully autonomous weapons cannot be relied upon to exercise the critical judgment that our highly trained, professional troops exhibit every day. They need to be deployed with proper guardrails, which don’t exist today.
To our knowledge, these two exceptions have not been a barrier to accelerating the adoption and use of our models within our armed forces to date.
The full statement is here.
“In a narrow set of cases”
Oh brother
I think this sums up the world (and its prospects): the billionaires are protecting democratic values against the fascists. This is almost on the level of a Monty Python sketch. If only it wasn’t true…
I’m not sure how much daylight can be seen between “the billionaires” and “the fascists”.
None. That is just the point.
If you can’t see the daylight between billionaires and fascists, you should probably see an optometrist.
or a taxonomist
Although the claim is true of some particular cases, historically and today, the generalization is at best inaccurate and at worst disingenuous. If there was no daylight, we’d expect the interests of fascists and billionaires to more systematically align, which they sometimes do but often don’t.
Well, I don’t know, but for me capitalism (certainly today’s late capitalism dominated by giant companies) and fascism ultimately go hand in hand. Sure, I wouldn’t say capitalists want fascism, but I also wouldn’t think they have much against it. Does this mean that their interests systematically align? A lot depends on what is meant by ‘systematically’ in this context. Their is no congruence of interests but it is clear, also from history, that if capitalists have to choose from among rival more radical approaches, they side with some form of autocracy. I do think they prefer liberal democracy (certainly the neoliberal kind) but that does not seem to be any longer on the table.
For almost any social group you could say that they don’t ’have much against it’ if you cherry pick the evidence, from billionaires to the working class, from intellectuals to producers, and yet we wouldn’t say there’s no daylight between them and fascists. Statistically I’d wager you could find more exceptions to the purported generalization than instances of the generalization by a factor of ten.
I am not convinced, but we could leave it at that.
The resulting responsibility gap is quite convenient: if things go south, Hegseth can always point out that the AI-powered robot was acting autonomously. “So, if we truly wish to calibrate our intuitions, an analogy more apt […] would be the cartoon revolver from the movie Who Framed Roger Rabbit, where the six-chamber cylinder is loaded with little cowboy characters. Would we
equip police officers with those sorts of guns? And, if a bullet-person spontaneously decided to head in a direction of its own choosing—as happens in the movie—would we be justified in holding the officer responsible? Upon witnessing the cowboy bullets take matters into their own hands, the movie’s protagonist Eddie Valiant tosses the revolver to the curb. When it comes to AI, an argument may be made that we should do the same […].” (from https://philpapers.org/archive/CHAQPQ.pdf )
I know some commentators on this blog tend to dismiss any talk of AI coming from a so-called tech-bro (imprecise term) as hype, regardless of content, but I really hope that isn’t the reaction in this case.
The following lawyerly discussion (or maybe just a well-articulated warning) seems pretty good:
https://davesecrest.substack.com/p/when-safety-becomes-sedition?r=1dltx&utm_medium=ios&triedRedirect=true
Two bits from Secrest’s discussion:
“Every prior surveillance state operated with bottlenecks: human analysts, processing time, storage limitations, legal exposure for individual collection decisions. Those bottlenecks created friction. Friction slowed the worst applications of surveillance authority. It created windows for legal challenge, for political resistance, for the kind of institutional pushback that, historically, has sometimes been enough to constrain the worst impulses of concentrated power.
AI at scale eliminates every one of those bottlenecks simultaneously.”
“We are not watching a procurement dispute. We are watching the negotiation over whether the government can compel a private company to build the instrument of its own citizens’ subjugation, and punish that company for refusing.”
I don’t think anyone would doubt that these are powerful systems to be used and, evidently, abused. What is strange here, also in the article that you linked in, is the tone of appalled surprise (for want of a better word). Exactly what did Americans, including many of these companies who were shoveling the money into Trump’s campaign and now are busy rolling out the red carpet for him, think would happen if they elect a fascist (or autocrat or use your pet expression)? That he will exercise restraint, respect constitutional rights and generally be a nice guy? Any powerful technology ever invented in human history has become, sooner or later, a tool of the state. What remains is then just self-restraint, which of course Trump and co have none of. Americans wanted this and now they are getting it. Unfortunately, we, who have nothing to do with the US, becoming collateral in this sad parody of a democracy.
Yes, you are right. Or at least, I agree with you. Still, I think it’s useful when anyone says clearly what is happening, even if they do it in a surprised tone.
Also – there is a difference here from same-old same-old processes in fascist, authoritarian etc. take-overs. And that is the power of this particular technology, and the failure of most people to fully grasp this (not to say I have a full grasp of it either, but I think most people really do not have any idea). But that last part is the part I am not allowed to say without those who worry about hyping AI landing and shouting in my ears.
Yes, I suppose there is a good chance that this is the end of American democracy. The question is what happens to the rest of us. I am, I must admit, more concerned about this. Right now, I have no answer. Positive, that is.
I don’t think this is marketing or propaganda, I think these statements reflect some genuine commitments. However, they must be read in light of the fact that Amodei is fiercely opposed to China, in general (and maybe partly out of those same concerns), but particularly in the race to AGI. It’s not clear to me that he would be so strongly opposed to ‘lawful but awful’ uses of Anthropic’s models to undermine China or even heat up the China-US trade (and one might fear other) wars.
With all that said, if you’re going to pick the side of one of the AI companies, you could do a lot worse than siding with Anthropic, and more power to them for doing something that’s not obviously in their own immediate interests.
Yes, but this is like choosing between two bad things (or more than two). The fact that this is where we are, and honestly, a lot of this if not most is thanks to the US, is the problem. We should not have allowed for-profit private companies to develop these technologies (this also includes social media of course). There should have been global coordination and oversight and massive regulation. Unfortunately, while I think Europe would have signed up to this (or at least I think we could have convinced the EU), the US model cannot accommodate anything of this sort. And so here we are: between a rock and a hard place.
> There should have been global coordination and oversight and massive regulation.
Given how this technology was developed this was absolutely never going to happen.
Choosing between two bad options is the stuff of life!
I really hope that this is not the case: there are instances when all options are bad, but that this is the stuff of life, I find this overly pessimistic.
I also don’t think that global oversight was not an option. As a state, you can at any point step in with control. Same with regulation on a global level. Of course, if you are a state that believes in lack of regulation, you won’t step in. And this is just what happened in the US, while the EU at least tries to regulate. However, it is hard to effectively control and regulate giant companies when they are headquartered in another state that does not want to regulate them.