Dem
47
GOP
53
The Pentagon wants to use Anthropic's chatbot, Claude, for military purposes. It has a $200 million contract with the company for using it. However, Anthropic CEO Dario Amodei wants some guarantees that the technology will not be used for autonomous weapons or mass surveillance of Americans. Secretary of Defense Pete Hegseth was not willing to give any guarantees and has said that the DoD can use Claude any way it damn well pleases and Anthropic has nothing to say about it. Amodei has said if he can't get the guarantees, he is not interested in the contract. Hegseth then said if Amodei doesn't cave by Friday, he will invoke the Cold-War-Era Defense Production Act, which can force any private company to supply the DoD with goods even against the company's will.
If it gets that far, it will be a very nasty court battle since that law was intended to give the government priority for buying scarce goods needed for national security. AI is not a scarce good. In fact, it is not really a good (product) at all. Also, the DPA only applies when there is a national emergency. The Supreme Court seems to be coming around to the idea that the president can declare a national emergency only when there is a, well, national emergency. At the moment, "Pete Hegseth is throwing a temper tantrum" does not constitute a national emergency.
One complication here is that the Pentagon wants to use AI for classified work. AI bots sort of need a kind of security clearance. Claude has already been approved. Elon Musk's Grok has also been approved, despite (or maybe on account of) its willingness to produce sexual content upon request.
The militaries in numerous countries see a role for AI in fighting wars, from letting drones pick their own targets to flying jet fighters. There are clear dangers in fully autonomous weapons since if they break the law, no one is responsible. Amodei understands this, but Hegseth doesn't care. Amodei has said that Claude can suffer hallucinations and cannot be trusted to make final targeting decisions since it could be wrong and cause unintended escalation or mission failures. He does not want to be on the hook for fatal decisions Claude makes, since he knows that is possible, and is warning Hegseth in advance. Hegseth doesn't care since he knows nothing about military strategy and even less about AI. But he does know he is the boss and everyone else (except Donald Trump) had damn well better kowtow to him.
Republicans generally oppose the use of government power to force companies to do things that they don't want to do, absent some very compelling emergency. If Hegseth does invoke the DPA, it will be interesting to see how Senate Republicans who don't like government overreach react. (V)