Anthropic refused to remove two safeguards from its AI models for the Pentagon: no surveillance of U.S. citizens and no autonomous weapons. The Department of Defense rejected the terms and threatened to blacklist the company. Meanwhile, the Pentagon signed a deal with Elon Musk's Grok instead, and AI systems in war simulations recommended nuclear strikes 95% of the time.
This is one of the most important storylines going on in the world of AI right now. If you're not familiar, up until recently, Anthropic was the only AI company that had access to classified information within the government. The US government actually used Anthropic's technology in the capture of Maduro in Venezuela. The US government got upset that they were asking about what they were using the technology for and basically said, we want to be able to use this for all lawful purposes. Anthropic essentially said, we're going to let you use our technology for whatever you want to use it for, with two exceptions. One, you cannot use it to surveil US citizens, and two, you cannot use it for autonomous weapons with no human in the loop. Pete hedges from the U.S. department of War basically said, no, you don't get to tell us what we can and can't use it for. Pete Hedgehog gave Anthropic until Friday to back down on the AI safeguards. They also threatened Anthropic to designate them as a supply chain risk, which is a big deal and has never been done to a US company before. It's usually done to foreign adversaries if they are designated as supply chain risk. That means that any company that has ties to the US Government is no longer allowed to use Anthropic either. On Thursday, Dario himself from Anthropic put out a statement. They reiterate in this article the two things that they do not want to relent on. But the government apparently doesn't have to look too far because, well, Musk's AI and the Pentagon reach a deal to use Grok in classified systems. So that thing I said earlier about Anthropic being the only company allowed to to operate with classified systems, well, now GROK is also getting that. And in completely unrelated news, this article came out from new scientists this week that says AIs can't stop recommending nuclear strikes in war game simulations in 95% of cases. Anyway, moving along.
AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know
