Meta reveals how Llama is powering US national security

Silicon Valley usually talks about open-source models in terms of developer ecosystems and open innovation. Meta just made it clear that those same models are plugging straight into military hardware and special ops control rooms.

Meta detailed how its Llama AI models are being used across the US national security sector. Partners like AWS and Snowflake are making Llama available within classified, highly sensitive government networks, while a crop of defense startups is already deploying it directly into the field.

Cutting report times by 18x in special operations

The most dramatic deployment comes from defense startup Legion Intelligence, which partnered with US Special Operations Command (USSOCOM) to launch SOFChat. It serves as USSOCOM’s first enterprise-wide generative AI platform, powered entirely by Llama.

According to Meta, SOFChat lets special operations forces generate intelligence reports 18 times faster and parse video footage 9 times quicker. In tactical environments where seconds matter, handing real-time ground monitoring and machine-speed analysis to soldiers is fundamentally altering operational tempo.

Because Llama’s weights are openly available, smaller defense startups can build specialized tools for military units without spending years reinventing the foundational model wheel.

Running offline on ordinary laptops

Frontline defense AI cannot rely on a constant cloud link. If comms cut out, the AI has to keep running.

Defense firm EdgeRunner AI adapted Llama to run completely offline on standard consumer laptops. That setup allows field units in disconnected, contested environments to identify safe landing zones for aircraft, translate foreign languages on the fly, and calculate precise rations for food and water without pinging a remote server.

Meanwhile, legacy defense prime Lockheed Martin is putting Llama to work inside its Skunk Works division. Software engineers hooked the model into their AI Factory platform to serve as a virtual subject-matter expert on the military's Joint Simulation Environment (JSE). The tool helps engineers write compatible code and build flight simulators for US and allied pilots, even when working outside their core expertise.

Why it matters

This is Meta’s most explicit push yet to prove that open-source AI is a vital national asset rather than a regulatory liability. Closed-source API providers have spent months in Washington arguing that releasing weights poses catastrophic security risks. Meta is countering that argument by embedding Llama directly into the Pentagon’s core workflow.

In high-stakes national security environments, having complete ownership of both the weights and the underlying data within air-gapped infrastructure is a strict prerequisite. Proprietary cloud models simply cannot play in many of these disconnected scenarios.

At the same time, we think this integration closes the distance between Silicon Valley code and kinetic warfare. As open models take on real-time battlefield intelligence, Meta will face uncomfortable questions about developer neutrality and the militarization of open software.