CARBONATO is described as the first botnet with an AI agent for command and control

A report from forkast.news describes CARBONATO as the first botnet of its kind: one whose command-and-control engine is an AI agent rather than a conventional server.

According to the report, the network has been running since October 2024. That means it was active for close to two years before it entered public discussion, and it puts AI agent and botnet infrastructure in the same sentence in a documented case for the first time.

Traditional botnet command and control relies on fixed servers, domain generation algorithms, or instructions written and dispatched by hand. Handing that layer to an AI agent means instructions can be generated, distributed, and adjusted by a model, which lowers the human effort needed to keep a network alive.

The significance here is less about the completeness of the technical detail than about the timeline it establishes. AI agents are no longer only assistants in demo videos; one has been sitting inside real attack infrastructure for a long stretch of time.

For defenders, the hard part is features. Fixed command-and-control domains and addresses can be blocklisted, and hand-written instruction flows have comparatively stable shapes. When a model writes the instructions, the space of possible traffic and behavior widens, and signature-based detection has to be rethought.

What to watch next is whether the disclosing party publishes a fuller technical analysis, covering infection scale, propagation, and the specific capabilities that were used, and whether security vendors can produce reproducible detection rules for model-driven instruction flows. Until then, CARBONATO is best read as a clear signal rather than a closed case.