This is no longer a chatbot on your laptop or a lab experiment in Silicon Valley. Artificial intelligence systems are already operating inside classified military networks, and the fight now is over whether they can be used for domestic mass surveillance and autonomous weapons under broad federal authority.
Anthropic has refused to expand the deployment terms of its Claude model to allow “all lawful use” inside secure defense systems. The Pentagon wants that authority. If the company does not agree, officials are prepared to terminate the partnership, designate it a “supply chain risk,” and have raised the possibility of invoking the Defense Production Act.
Pentagon spokesman Sean Parnell made the deadline unmistakable.
“They have until 5:01 PM ET on Friday to decide. Otherwise, we will terminate our partnership with Anthropic and deem them a supply chain risk for DOW.”
Claude is not sitting idle. Contractors have built workflows around it. Programs are integrating it. Once a system reaches that stage, whoever controls the deployment terms controls how far it runs inside surveillance pipelines and weapons platforms.
Anthropic has drawn its line at two points: no domestic mass surveillance and no fully autonomous weapons operating without meaningful human oversight. Those limits are already written into its defense agreements. The Pentagon wants broader language that covers “all lawful use” once Claude is embedded.
Dario Amodei has not budged.
“Those uses have never been included in our contracts with the Department of War, and we believe they should not be included now.”
Claude can rip through massive data streams in seconds. It can cross-reference databases, surface patterns across millions of entries, flag anomalies, and plug directly into targeting workflows. At scale, that is not incremental. It changes the speed and scope of what surveillance systems and weapons platforms can do.
Read More: Oops: Amazon’s AI ‘Deleted and Recreated’ a Live AWS Environment
Drop that inside domestic data systems, and it can map associations, movements, communications trails, and behavioral signals at a pace no human team could match and at a scale no oversight body could realistically audit in real time. Drop it into weapons infrastructure, and it can help sort, rank, and prioritize targets before a human operator signs off, compressing decisions that once took hours into moments and narrowing the window for human hesitation.
The Pentagon wants the authority to determine how far those capabilities go once the system is inside secure networks. Anthropic is refusing to hand over that discretion.
Amodei has warned about where that road can end.
“In a narrow set of cases, we believe AI can undermine, rather than defend, democratic values.”
Once systems like this are embedded inside classified networks, they do not sit idle. They ingest data continuously. They surface patterns across entire populations. They accelerate decisions that ripple outward through surveillance systems and weapons platforms. Their outputs become inputs for the next cycle.
Claude is already inside the machine. So is Grok.
The limits set today will govern how those systems operate at scale tomorrow. Once those limits are defined under a broad federal interpretation of lawful use, they become the operating baseline.
Baselines in systems like this rarely shrink.
And once they harden, they do not unwind. They expand quietly.
Editor's Note: With President Trump back in the White House, the state of our Union is strong once again.
Support RedState’s coverage of the president's State of the Union Address and help us report the truth the radical Left doesn't want you to hear. Join RedState VIP and use promo code POTUS47 to get 74% off your VIP membership.
Join the conversation as a VIP Member