Pentagon signs AI deals with six firms after talks fail with Anthropic

In short: The US Defense Department has signed AI agreements with six technology companies after negotiations collapsed with Anthropic over how the Pentagon could use its Claude system.

What happened

The Defense Department had previously awarded Anthropic a large contract to run its Claude AI model on classified networks, which are government computer systems used for secret information. Those talks broke down by a deadline, after Anthropic refused to allow the Pentagon to use Claude for “any lawful purpose.”

Anthropic said it was worried the wording could open the door to mass surveillance of Americans and to weapons that act on their own. The Pentagon said it offered written assurances that existing rules would still apply, including limits on domestic spying and requirements for people to stay in control of targeting decisions. Anthropic said the promises were not strong enough and could be worked around.

After the split, the Defense Department cut ties and labeled Anthropic a “supply chain risk,” meaning vendors that sell to the Pentagon are not supposed to use Anthropic in their work. The Trump administration also ordered federal agencies to phase out Anthropic technology within six months. Anthropic has sued the Pentagon to challenge the label.

The Pentagon has now secured agreements with six AI providers, including OpenAI and Google, that allow “any lawful purpose” use. Reports also name Meta, xAI, and Palantir, with a sixth company not publicly confirmed.

Why it matters

This fight helps set the rules for how the government buys and uses AI. If “any lawful purpose” becomes the standard, companies may face a choice between accepting broad government use, or losing access to major public contracts.

Source: NYTimes