When the DoD Meets the DAO: A Clash of Clauses, Egos, and 'Lawful' AI

Michael C. Horowitz – a political-science prof at UPenn and a former Pentagon deputy – suggests the whole Anthropic-DoD drama is less about fine print and more about a classic case of personalities and politics cosplaying as a contract dispute. It’s the bureaucratic equivalent of two whales arguing over slippage while the pool drains.

By signing up for classified national security work, Anthropic effectively minted itself as the first frontier AI lab to become a government “vendor,” securing what might be the world’s most secretive smart contract. They got their “I did a classified” merit badge, but now they’re reading the terms and conditions with growing alarm.

The Pentagon’s new AI policy mandates an “all lawful uses” clause in every vendor contract, treating advanced AI like a weapon system you can just add to cart. To Anthropic, this feels less like a procurement strategy and more like the DoD is FOMO-buying capabilities with the ethical safety off—a bit too eager to deploy before mainnet.

Anthropic’s brass is spooked that runaway AI could supercharge mass surveillance, even domestically. Horowitz points out that blaming the Pentagon alone for this fear might be a mis-targeted airdrop; the surveillance state, after all, has many validators.

When it comes to autonomous weapons, Anthropic isn’t philosophically opposed—it just insists its tech isn’t battle-hardened enough for a kill-switch. The company’s stance is essentially to HODL its models until they pass a rigorous audit, rather than aping into a deployment that could spawn a rogue agent on the battlefield.

In reality, Anthropic’s Claude is already a node in the military’s decision-making stack, feeding data into systems like Project Maven to give commanders a better read. Think of it as a high-stakes oracle network providing price feeds, but the “asset” is real-time intel and the “exchange” is an active warzone.

This whole showdown highlights how tech-government partnerships are shaped by both hard regulatory forks—like the “all lawful uses” clause—and the very human consensus mechanisms of ego and politics. As AI gets more deeply integrated into strategic ops, the ethical debates around privacy, surveillance, and autonomous weapons are only getting more convoluted, like a governance proposal with infinite amendments.