Selling a pen labeled "this pen will refuse to sign certain orders" is not sabotage or malicious and is thus not a supply chain risk. The DoD is free to not buy from Anthropic, but designating them as a supply chain risk is incorrect, as well as arguably arbitrary and capricious given their public criticism of Anthropic's beliefs.
I don't think anyone doubts that subcontracts exist. They doubt that Anthropic's insistence that Claude isn't capable of being the operating system for an automated killbot and associated terms the DoD previously agreed to means that there's any danger to the US military from Lockheed Martin using Claude as a code assistant to build GUIs.
What if it’s a GUI to manage deployment of the kill bots? I worked for a DOD contractor on software. All our scenarios were more related. It was like “OK you’ve overthrown the civilian government of this Middle Eastern country, including the radio regulators, and now we need to coordinate radio frequency assignments for the tactical squads.”
Code which renders dots on a map isn't going to magically reorder itself if the GNSS endpoints get switched from manned to unmanned assets just because it was authored by Claude, and if the subcontractor isn't auditing and testing the code outputs properly then they're the supply chain risk, not a specific code-gen tool they use. Obviously if instead of being used as code or report generating assistants, LLM agents are used to execute combat decisions in real time that introduces a lot more risk. That's... literally Anthropic's argument against using them for that specific purpose.
It doesn't make much sense to determine that because Anthropic continues to insist that Claude is unsuitable for use as some sort of super advanced loitering munition, Lockheed using it to write job ads is as much a threat to national security as a Chinese comms terminal.
It's not like the DoD and their subcontractors and their subcontractors' subcontractors aren't used to all sorts of fine-grained separation when it comes to what is and isn't classified or export controlled under normal circumstances, or that Hegseth taking a break from fantasising about soldiers' testosterone levels for long enough to decry Anthropic's stance as 'woke' makes the decision look like one based on operational effectiveness.
"Supply chain risk" specifically refers to risk of malicious attack or sabotage through the supply chain, not all risks associated with supply chains.
A vendor with particularly poor QA might be a risk in your supply chain but isn't a "supply chain risk" according to this designation and neither would Anthropic be.
Supply chain risks are usually when someone does a attack over the supply chain. Like when the mossad planted literally explosives in radio devices for Hezbollah.
Antrophic merely said they don't allow their tools to be used for autonomous killing.
A very specific case.
So where is the danger making it necessary preventing all suppliers from using Claude at all? That only concerns those who work directly on autonomous killing and only they will have to go to Altman or Musk.
So none of that is any threat to the military - so it was just black mailing to make them do the governments will - which is something usually dictatorships do, not democratic systems.
In a democratic society, the default should be, the military just buy somewhere else, then force companies to build them the terminator. So no private company controls what the military does, they just control their product. You want the military to control private companies? There is a name for that kind of system, where this is conmon.
Eh this just we wont buy from you because you are putting limitations on the use. Anthropic will be fine if they don’t like it they can accept the terms of the contract or walk away.
Why has this never previously been done to an American company from which the DoD decided not to purchase something?
Perhaps people who are noticing politics here - including in the decision of two Trump-appointed judges, against the he vote of another Republican-appointed judge- are not the ones who are choosing not to see reality?
President Truman literally seized the assets of a steel company to avert a strike that would hurt the Korean War effort. The Supreme Court stopped him, but three Justices voted it was okay!
We also have the Defense Production Act: https://www.congress.gov/crs-product/R43767. It doesn’t happen more often because most people aren’t stupid enough to say “no” to the U.S. military. These Effective Altruists, however, are too big for their britches.
I don't know, I am just guessing because not a single American company has publicly stated they are against its product use up to a certain degree by DoD? If they have not publicly declared it then yes, DoD simply don't sign the contract.
So your argument is that the government is punishing Anthropic for its public statements? And it is correct to do so?
(I think it's also the case that Anthropic didn't make any public statements about the negotiations until after the DoD was already publicly threatening this designation. But that seems less important than the question of: is it good for the US government to try to destroy companies whose public statements it dislikes?)
Honestly I would find a certain satisfaction in a world where the next Democratic administration takes away Fox broadcast licenses, forbids use of Oracle by any government contractor unless Ellison hands CBS over to Rachel Maddow, and tells Bezos that Amazon is going to be banned from public roads unless the Washington Post front page has positive stories every day. But I doubt the same is true for the many pro-Trump people who complained vociferously about Biden officials expressing preferences to social media companies because of the merely implicit threat that they might take government action of some kind if they weren't listened to.
So we have a pen that may refuse to sign certain orders, and is clearly labeled as such.
If the DoD needs to sign such orders, they won't buy this pen.
If a supplier needs to sign such orders, the supplier also won't buy the pen.
If a supplier needs to sign other orders, and this pen is the best one available, the supplier may decide to buy this pen. Since the orders are within the range of what the pen will sign, they get signed, and the supplier fulfills its contract with the DoD.
Where is the supply chain risk? The ink isn't going to erase itself post-hoc.
---
Now, if the DoD suspects that Anthropic will train its models to try to actively sabotage military operations, that's a very different story. Does anyone actually believe that?
> If a supplier needs to sign other orders, and this pen is the best one available, the supplier may decide to buy this pen. Since the orders are within the range of what the pen will sign
Who decides what’s within the scope of “what orders are within the range of what the pen will sign?” The boundaries are never clear cut. You’ve got a vendor who could kill switch critical military technology if it’s used in a way they decide falls outside the scope of what they want.
> You’ve got a vendor who could kill switch critical military technology if it’s used in a way they decide falls outside the scope of what they want.
If the DoD hires ACME to build some software to make widgets, and ACME uses Claude to build that software, where is the kill switch?
I guess the Claude model could hide a kill switch in the widget manufacturing code, but so could ACME's human subcontractor. Why is Anthropic being considered a supply chain risk here?
The model can refuse to build something, but it can't take back what has already been built.
If you outsource your thinking to a server that can detect your IP address, and decide to render decision “b” instead of decision “a” to your query, that’s a risk. (Or detect by the type of query, past history of your use, etc.). Panopticlick should be having a field day with all this.
Oddly when the topic of e.g. car infotainment systems comes up, like BMW locking you out of features in your own car, this place is up in arms, but when the military has a problem with it, suddenly they should be forced to accept it?
That isn't sabotage, that is simply product design. Anthropic is free to create their products that refuse to kill people, and the DoD is free not to buy those products if they don't fit their needs.
However, that's not what a supply chain risk is. It's not an order to "don't buy these products if they don't work the way we want them to", its designating Anthropic as a national security threat because they might intentionally sabotage US military operations.
As I said elsewhere, it's utterly preposterous that the DOD actually considers this a reasonable threat, because it's simply not a reasonable possibility. There's no way Anthropic would do that, precisely because of the consequences that would follow if they did, and got found out. Moreover, they already clearly stated their terms and preferences. It's all out of the open. There's no supply chain risk, that designation is purely political.
I work in the space industry and regularly use parts in our designs that are unapproved from space flight. Sometimes it's because it's a commercial part that we seem acceptable for flight, other times it's an unscreened or engineering model that doesn't go through the proper testing that space-qualified parts do. One vendor even dents the lid of a part to invalidate the hermetic seal guarantee that they claim for the space-qualified version (it still works fine). Many will have fine print in the data sheets saying the part is not to be used for critical applications like aerospace or medical devices. Sometimes we tell a little fib to our vendor that we are just developing non-flight devices so that they don't get upset and refuse to sell the lower-grade part to us. These vendors are scared of having something fail in an unapproved environment and are unaware of how much risk the customer is tolerating by the use of these unscreened parts.
It would have been incredibly simple for Anthropic to say "we cannot guarantee performance in a kill-chain application due to an unknown level of safeguards implemented in the baseline product and cannot estimate a cost for developing a new product capable of such application." and leave it at that.
There are plenty of companies that are opposed to certain operations of the US military and refuse to sell to the US military on those grounds. None of them are considered a supply chain risk because that's not what a supply chain risk is.
A supply chain risk is specifically an entity that is likely to maliciously sabotage the US military's operations, such as Chinese companies that are controlled by the CCP. It is not simply a decision to "not buy from this company", but rather a wide-ranging decision that this particular company is a national security threat.
Or is known to have weaknesses that might permit such actions. Microsoft permitting Chinese workers to maintain DOD systems presented a supply chain risk, even though MS themselves were not the ones likely to behave maliciously. Instead their policies were so stupid that they invited potential malicious actors into the facilities.
Would you care if your accountant liked to sit outside a high school and watch teenage girls with binoculars? What if a local shop owner was known to harass female employees? Everybody has a line somewhere.
I guess it depends on how to measure a single "person"? If you spun up 2000 copies of Terrance Tao, I wouldn't be surprised if you found a few new discoveries at the end of it.
I just don’t think it’s true, or is significant enough to matter to the direction of travel of AI even if there were something to it. It’s another cope post being lobbed at the idea of AI going somewhere and I’m sick of them.
This reminds me a lot of the proof by construction for the 4-color theorem. It was only enabled by the advancement of computers and dissatisfied many of the computer scientists and mathematicians since it was a "brute force" approach.
I wonder if AI will end up being similar. Certain theorems get proven by AI but others do not. We haven't reached the limits of this yet and I haven't found a good argument for where those limits will be (I do doubt that there are no limits).
> I’m surprised there isn’t a Costco like medical group that’s nationwide, has a membership, and works solely to provide care efficiently.
What you are describing is an HMO, which hasn't had that much lower costs historically. Theoretically, you pay once and then they take care of you, but in practice costs haven't been that much lower.
reply