The US Navy plans to deploy LLMs this year without any internet connection.

On July 2026, US Chief of Naval Operations, Hong Cao released the “Arm with Data and AI” strategy document which came into effect right away. It allows the use of large language models and agentic AI to run on ships and Marines without any Internet connection. The focus is on reducing the Mean Time to Effect (MTTE) i.e. compressing the timeline between data gathering and taking action on the ground.

This strategy is based on the “Bits2Effects Cycle” which is a 5 step process starting from gathering data on the battlefield to taking action in the real world. There’s also an AI War Council responsible for prioritizing applications and they have the power to modify data sharing policies on the fly. In the last 6 months, the military’s own LLM platform GenAI.mil went from 80,000 to 1.5 million daily active users. By the end of this year, they plan to have some core metrics in place and the goal is to double the number of data engineers by 2029.

But the most interesting part of this document is that it explicitly states that “Delaying AI deployment is considered a bigger threat than ‘imperfect alignment’ (i.e., model mistakes and hallucinations). This is the first time that the military has put in writing that delay is worse than having an LLM that has access to your combat systems, and that speed trumps accuracy.”