Navy Declares Itself 'AI-First,' Plans to Run Language Models Directly on Warships

The Department of the Navy has signed a formal strategy to weaponize data and artificial intelligence, laying out a plan to build an "AI-first" fleet speed — even when that means accepting imperfect safety guardrails.

Acting Secretary Cao framed the strategy as a roadmap for out-learning and out-fighting adversaries. At its core is a five-stage framework called the "Bits2Effects Cycle," which traces the path from automated data collection through classification and analysis to real military action, with lessons fed back into the loop. A metric called "Mean Time to Effect" — the window from capturing new data to producing a concrete response — is the yardstick. The shorter that interval, the faster a force adapts. In a prolonged conflict, the side with the fastest learning cycles wins.

The strategy sets aggressive timelines. Most measures are supposed to be in place by the end of calendar year 2026, and the Navy wants to double its count of qualified data engineers, data scientists, and AI/ML engineers by fiscal year 2029. Six stated goals include accelerating operational AI deployment, improving data usability, expanding technical infrastructure, streamlining approval workflows, boosting AI literacy among personnel, and tightening collaboration with industry, academia, and allies.

The operational vision goes well beyond office tools. The Navy plans to run large language models and agentic AI directly on warships and with Marine Corps expeditionary units — systems that must function even when comms are jammed or severed. Service members would build their own applications on top of these models. An "AI War Council" would prioritize use cases, coordinate resources, and pre-approve changes to data-sharing and deployment rules for wartime conditions.

Crucially, the strategy paper adopts a far-reaching trade-off from the Pentagon's broader AI doctrine: the risks of moving too slowly outweigh the risks of imperfect alignment. Within what it calls a "Wartime Approach," the department wants to handle risk and organizational obstacles as if the country were already in an active conflict, privileging speed over caution.

That urgency is grounded in a landscape that is already shifting fast. GenAI.mil, the Pentagon's central generative AI platform, hit 1.5 million daily users in June 2026 — up from 80,000 at its December 2025 launch. The US military reportedly used Anthropic's Claude for target analysis and strike planning during operations against Iran. A Navy AI program is said to have cut a submarine planning task from 160 hours to ten minutes. Meanwhile, China's PLA is testing AI for unmanned combat vehicles, cyber defense, and target acquisition, and NATO allies are using the technology to track Russia's shadow tanker fleet.

The strategy is likely to further intensify demand for powerful language models from both Anthropic and OpenAI, whose deals with the Pentagon have been shaped by politically charged debates over autonomous weapons and data safeguards. The Navy is staking its future on the bet that the force that learns fastest wins — and that hesitation is the costlier option.