When business ethics and the military's view on national security come into conflict.

I'm David Brancaccio in Los Angeles.

The artificial intelligence company Anthropic is loosening some of its core safety principles.

This at the same time, the company faces pressure from the Pentagon to rollback limitations on how Anthropics' Claude AI models are used.

Marketplace's Nancy Marshall-Genzor is here now with some details.

Well, David, Anthropic unveiled a new policy on safeguards earlier this week,

and it's moved from self-imposed guardrails to non-binding goals for AI safety.

In a blog post on Tuesday, the company said under its old policy,

if Claude became capable of, say, helping build a weapon, Anthropic would adopt new stricter safeguards,

and it hoped other companies would do the same, and governments would coordinate with it on this,

and that just did not happen.

Now, there's also pressure from the administration

on related matters. What's the Pentagon's concern? Well, there are reports that Defense

Secretary Pete Hegseth has given Anthropic an ultimatum, and Hegseth wants the company to roll back

its rules even more by tomorrow, or it could lose a Defense Department contract worth $200 million.

The Pentagon doesn't want any constraints on AI use and weapons. For example, if it has just

minutes to fire weapons and needs AI to do it, it doesn't want to have to ask Anthropic for

permission first. But Anthropic wants to be sure Claude isn't used for

things like government surveillance or autonomous weapons.

And HegSeth has other tools to pressure Anthropic via what its business partners?

Axios is reporting the Defense Department could designate Anthropic as a supply chain risk.

As a first step in that process,

...