Palantir shows how the military could use AI chatbots to plan operations

Palantir Technologies has held a series of closed-door demonstrations showing how the military can use large language models, Anthropic's Claude in particular, to analyze intelligence data and suggest operational plans. Materials from these demos, along with Pentagon documents reviewed by Wired reporters, offer the first public look at how AI chatbots are being integrated into military command systems.

Palantir, which specializes in data analytics for government agencies and intelligence services, has long worked with the US Department of Defense. But this is the first time it has publicly demonstrated chatbot interfaces built specifically for military planning. In the demos, the AI system took in intelligence data, analyzed it, and proposed concrete next steps to officers, ranging from logistics decisions to tactical recommendations.

The key detail: some of the demos used Claude as the language model. Anthropic markets itself as a developer that puts AI safety first, which makes this awkward. The company publicly emphasizes responsible use of its technology, yet its models end up powering military planning systems. Anthropic is not the first civilian AI vendor to work with the defense establishment, but the nature and scale of this integration have sparked a fresh round of debate across the industry.

The Pentagon documents make clear this is not a paper exercise or a pilot project. The military is actively working out real combat applications. LLM-based systems are being considered as tools to speed up the so-called OODA loop: Observe, Orient, Decide, Act. In military operations, the speed of processing intelligence and making tactical decisions is critical, and AI assistants could provide a tangible advantage here.

Palantir pitches its AIP (Artificial Intelligence Platform) to government customers as a solution built around controllability and data security. In the demos, an operator asked the system questions in natural language, and the system pulled data from multiple sources and returned a structured answer with recommendations. According to the company's official position, the final decision always rests with a human.

Critics, however, point to the risks. When an AI system suggests next steps in a combat setting, the line between recommendation and decision blurs. Cognitive research shows that people tend to trust automated systems even when their conclusions are wrong, especially under the stress and heavy information load that military commanders typically face. It's a serious problem the industry has not yet solved.

Data is a separate issue. The intelligence these systems process falls under the highest classification levels. Palantir holds the necessary clearances and has years of experience working with classified infrastructure, but integrating commercial LLMs raises new questions about security, control over the models, and potential vulnerabilities.

All of this is part of a broader trend. The militaries of major powers are actively exploring how AI can speed up command and control. The gap between public pledges of responsible AI and how the technology is actually used in military systems is becoming increasingly obvious.

For the industry, the signal is clear: companies building large language models inevitably become part of the defense technology stack, whatever values they profess.