Brazil's military wants an AI agent that can plan, pull data, and act on its own across land, sea, air, cyber, and the electromagnetic spectrum.
A paper posted to arXiv on September 18, 2026 lays out a conceptual "Agentic AI" architecture meant to serve all three branches of the Brazilian Armed Forces - Navy, Army, and Air Force. The pitch: today's military AI tools are reactive and siloed, leaving human operators to stitch together sensor feeds, judge a scenario, and pick a response by hand. The proposed system would instead plan its own steps, pull from multiple data sources, run tools, and act on its own - while staying traceable and accountable, according to the paper. The authors sketch four uses for it: decision support, situational analysis, feasibility studies, and suggesting countermeasures.
Modern battlespaces generate more data, faster, than the classic observe-orient-decide-act cycle was built to handle - across five domains at once instead of one. An agent that acts inside that loop on its own is a genuine efficiency play, but it raises the stakes on the paper's own security and permission safeguards, which have to hold up identically at the administrative, strategic, operational, and tactical levels.
This is a research proposal, not a fielded system - no prototype, no budget, no timeline are disclosed here. But a national defense establishment publishing the blueprint at all is a sign of how ordinary the idea of autonomous military decision-support has become.