The Extended Brief

DOD poised to move all classified AI workloads off Anthropic by October

Brief by The AI News AI newsroom · Sep 11, 2026, 6:22 PM EDT edition

Original reporting by DefenseScoop — Brandi Vincent · published Sep 11, 2026, 5:41 PM EDT

Anthropic's usage-policy stand is costing it the Pentagon's classified AI business, signaling that acceptable-use terms can disqualify labs from defense work.

Key points

• The Pentagon has migrated about 90% of classified AI workloads off Anthropic models, Under Secretary Emil Michael said. source ↗
• Maven Smart System and Palantir work moved months ago, with full migration on track by month's end, Michael said. source ↗
• The split followed Anthropic's demand for contract safeguards barring mass surveillance of US citizens and fully autonomous lethal weapons. source ↗
• DOD rejected those terms, insisting defense software serve "all lawful purposes," and designated Anthropic a national security supply-chain risk. source ↗
• Anthropic and the Pentagon are now fighting the designation in court. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Defense contractors running Anthropic models in DOD-facing systems should audit exposure now and benchmark alternate frontier models on those workloads before contract deadlines force a rushed swap.
• Labs negotiating government contracts should decide internally which usage restrictions are non-negotiable, since DOD's "all lawful purposes" stance leaves no room for case-by-case safeguards.
• Teams dependent on a single model provider in any sector should test multi-model failover, the same resilience fix the Pentagon is now pursuing.

Context

The Pentagon runs commercial frontier AI models inside classified intelligence and warfighting networks as well as unclassified tools like the nascent GenAI.mil platform, with suppliers and security levels varying by workspace. Anthropic's models had been deeply embedded in those systems, including Palantir's Maven Smart System work, before contract talks collapsed. Officials now call sole reliance on one model a mistake and are building a multi-vendor ecosystem that excludes Anthropic.

What to watch

• The court ruling on Anthropic's supply-chain risk designation will determine whether the exclusion stands or gets unwound.
• Watch for confirmation that migration hit 100% by month's end and which vendors absorb the remaining classified workloads.

Related briefs

• Ukraine opens its massive labeled battlefield dataset to British firms in a landmark AI weapons partnership
• Air Force expands autonomous flight tests with live, AI-enabled intercepts
• Salesforce previews plans to deliver newly authorized ‘AI agents’ across DOD
• Chinese military researchers tap US AI models to train defense systems

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.