Rush AI into combat
Frontier AI rushed into combat systems. The Pentagon rushes frontier AI into targeting, weapons and command systems, with fewer checks.
What it does
Frontier models go quickly into targeting, weapons and command systems, with fewer review steps and pressure on vendors to allow 'all lawful uses'. Vendors that insist on use limits risk losing contracts. Back-office uses like logistics and paperwork aren't what this law is about.
Applies to: Combat, targeting and command systems at the Pentagon and intelligence agencies.
- The Pentagon puts frontier AI into targeting, weapons and command systems as fast as possible, with fewer review steps.
- Vendors are pressed to allow all lawful uses; those that insist on limits risk losing contracts.
- Back-office uses like logistics and paperwork aren't covered: they aren't what raises the risk.
Where things stand
GenAI.mil (launched Dec 9, 2025) gives DoD staff access to Gemini, ChatGPT and Grok, with over 1.2M users by April 2026. OpenAI signed a classified-network deal on Feb 28, 2026 that it says keeps bans on mass surveillance and requires human responsibility for force.
Why it gains ~4 days
Big contracts for combat systems bring labs money and demand: about a day and a half gained a year.
Biggest unknown: Whether speed-first deployment with weaker vendor use limits produces real capability gains faster than it produces accidents, misuse, or loss of top labs' cooperation.
Why it raises p(doom) by ~0.15%
Not a safety law. Puts frontier AI into weapons, targeting and command as fast as possible.
Racing AI into weapons and command systems raises the stakes of accidents and escalation, and pushes China to do the same.
The strongest case that it costs more
Supporters argue the military that adopts AI first deters war, and that falling behind China is the bigger danger.
The debate
For
- Defense Secretary Pete Hegseth Jan 2026: told DoD to 'accelerate like hell' and said 'We will not employ AI models that won't allow you to fight wars.'
- President Trump (NSPM-11) June 2026: directs agencies to 'accelerate AI adoption' and eliminate 'unnecessary barriers to rapid deployment'.
- Under Secretary / CTO Emil Michael 2026: AI strategy requires DoD to match the speed of the private AI sector.
- Michael C. Horowitz and Vinh Nguyen (CFR) June 2026: credit NSPM-11 with chain-of-command accountability and assurance infrastructure, while flagging gaps.
Against
- Sen. Chris Coons (D-DE) Feb 27, 2026: defended Anthropic's limits on mass surveillance and autonomous weapons as 'commonsense policies'.
- Sen. Ruben Gallego (D-AZ) June 2026 letter: warned rushed changes risk friendly fire and civilian harm.
- Anthropic 2026: refused Pentagon demand to drop limits on fully autonomous lethal weapons and domestic mass surveillance; is contesting the supply chain risk label in court.
Sources
- DefenseScoop: Pentagon awards frontier AI contracts (July 14, 2025): Four $200M-ceiling CDAO awards.
- NSPM-11 (White House, June 5, 2026): Sec. 2(a) adoption; Sec. 3(b) terminate vendors inconsistent with policy; Sec. 3(f) rescinds NSM-25.
- Military.com: 1.2M using GenAI.mil: User and agent counts, model lineup.
- Al Jazeera: US court upholds Pentagon's blacklisting of Anthropic (Sept 25, 2026): Latest ruling; conflicts with an Aug 2026 district court ruling.
Rough starting points, not precise forecasts. Lead costs assume China doesn't depend on U.S. models, the case least favorable to safety laws, and count 3 years. On the menu you can change every assumption and put in your own numbers. Last priced 2026-09-26.