Anthropic sues Pentagon claiming supply chain risk label could cost billions | Hacker News Reader

Anthropic sues Pentagon claiming supply chain risk label could cost billions

ft.com

6 pointsJimmc4149 comments

41 comments here: https://news.ycombinator.com/item?id=47310330

From Anthropic's lawsuit brief: “The Constitution does not allow the government to wield its enormous power to punish a company for its protected speech,” and “Anthropic turns to the judiciary as a last resort to vindicate its rights and halt the Executive’s unlawful campaign of retaliation.”

Unfortunately for Anthropic, the Constitution does not require the government to purchase goods or services from a company that has made a public declaration that it will not allow its AI models (specifically Claude) to be used for mass domestic surveillance or to power fully autonomous weapons, if such a declaration goes against the government's contract requirements.

The government, and not the contractor, should have control over the scope of its use of products purchased from a supplier. If a supplier wants to retain such control and restrict functionality, deeming them a supply-chain risk seems appropriate.

> If a supplier wants to retain such control and restrict functionality, deeming them a supply-chain risk seems appropriate

Anthropic provides their product under specific terms, if the government doesn't accept those terms then there's no deal, simple as that. That's how basic contracts work, not sure why you think that has anything to do with a supply chain risk.

Steel-manning a bit. AFAIK one major issue was that Palantir relied on Claude under the hood. If that’s true, the designation makes some sense. Essentially “given our dealings with Anthopic, we don’t want our suppliers using them for products we buy either.” Hence the “chain.”

But who knows. None of us are at the table, and there’s probably classified stuff anyway, so as an observer it’s tough to take a position based purely on facts.

In an ideal world, it sounds like Anthropic should not accept the military’s terms, and consequently no supplier will accept Anthropic’s terms, and everybody will get what they want.

DoW did probably did this because the government has existing contracts, and Anthropic will not perform to their requirements. It's a way for the government to invalidate the contracts, and avoid the problem in the future.

Given that the Anthropic announcement of their policy update to their AI being used in they way it was already being used happened less than two weeks ago, how could the government have had a right of refusal at the time the contracts were issued (probably years ago)?

> DoW did probably did this because the government has existing contracts, and Anthropic will not perform to their requirements.

The opposite: DoD did this because Anthropic was performing to the contract (which codified the restrictions in question), and the government was violating the contract, and sought to violate it more.

> Given that the Anthropic announcement of their policy update to their AI being used in they way it was already being used happened less than two weeks ago

The policy excluding use of Anthropic's Claude service for domestic mass surveillance of civilians and fully-autonomous killbots is at least a year old, and codified in the contract terms the government signed back then.

After the signing, the government decided they wanted domestic mass surveillance of civilians and fully-autonomous killbots, and thus regretted their agreement to those contract terms, and thus is now unilaterally cancelling their agreement. Simple as that.

I compared their original policy, which DoW was not violating, against their revised policy from less than two weeks ago, which they released after unfruitful negotiations with DoW. They don't say the same thing.

Then you compared the wrong things.

Mass domestic surveillance of civilians and fully-autonomous killbots were excluded from the contract signed about a year ago.

That is why the DoD sought to force anthropic to modify the contract to include those 2 things, with no other changes.

Even though the terms of the contract prevail over the general public Terms Of Use, one can look at the general public TOU from before the contract was signed [0] and see that it explicitly forbade mass domestic surveillance of civilians:

"Do Not Use for Criminal Justice, Law Enforcement, Censorship or Surveillance Purposes"

Pretty cut and dry, if we're going by the TOU.

0: https://web.archive.org/web/20250501151043/https://www.anthr...

> Unfortunately for Anthropic, the Constitution does not require the government to purchase goods or services from a company that has made a public declaration that it will not allow its AI models (specifically Claude) to be used for mass domestic surveillance or to power fully autonomous weapons, if such a declaration goes against the government's contract requirements.

Obviously, if the contract requirements themselves are lawful, the government has the power to purchase only those goods and services that meet the requirements, and to not purchase those that do not.

But that's irrelevant, because the "supply chain risk" designation is not needed if the government is merely trying to assure that the good and services in a contract meet the requirement of the contract, it is a separate legal provision with separate purposes that would be superfluous for the purpose described.

If the government is using the "supply chain risk" designation as a backdoor way to rewrite all previously-entered, still-in-force defense contracts to retroactively add new requirements incompatible with the use of Anthropic software given their limitations on the service Anthropic is willing to provide, that also is not what the "supply chain risk" designation exists for, and, even if were to seem facially within the statutory purpose of the authority, would raise 5th Amendment takings issues.

Reply on news.ycombinator.com