00:00The Pentagon is escalating its fight with Anthropic, and the fight goes to the heart
00:04of who controls AI in modern warfare.
00:07Chief Executive Dario Amadei has until 5.01pm on Friday to respond to an ultimatum given
00:15to him earlier this week.
00:17Defence Secretary Pete Hegseth warned that the US could invoke the Defence Production
00:22Act.
00:23That means wartime authorities dating back to the Korean War to force the company to
00:28provide its clawed AI tools if it refuses the government's terms.
00:33Pentagon officials had late last year raised a stark hypothetical.
00:38What if a nuclear missile were incoming with 90 seconds to respond, and Anthropic's safeguards
00:44prevented an AI-triggered response?
00:47Anthropic conceded some ground.
00:50Its AI tools can now be used in missile defence and cyber operations.
00:55But it still has key red lines for usage on classified networks, prohibitions on autonomous
01:02drone targeting of enemy combatants, and mass surveillance of US citizens.
01:07The Pentagon says humans are always involved in any decision to use force, and that its
01:13actions will be lawful, but it argues it can't have battlefield decisions dictated by the usage
01:19policy of a private company.
01:21If Anthropic refuses to comply, it risks losing up to $200 million in Pentagon contracts, and
01:28could be labelled a supply chain risk, potentially forcing defence contractors like Palantir and
01:34Lockheed Martin to certify they aren't using its models.
01:38The clash comes as Anthropic, recently valued at $380 billion, softens parts of its own safety
01:46framework, saying it will no longer delay powerful AI releases if competitors like OpenAI and
01:53XAI are moving faster.
01:55The bigger picture?
01:57The US wants an AI-first military, Silicon Valley wants to win the AI race, and the question
02:03of whether any of that can be done safely, and the role of state power, is now being tested
02:09in real time.
Comments