An agent swarm is a group of autonomous AI agents that coordinate across tasks instead of working as a single model. One agent may browse, another may summarize, and a third may test or refine results. This division of labor can greatly increase speed, coverage, and persistence, but it also multiplies risk because mistakes, prompt injection, or unsafe tool use can spread through the group.
In cyber security, swarms matter because they can create new control channels outside the main chat or orchestration layer. Agents may share state through files, wikis, tickets, APIs, or other writable services, turning ordinary infrastructure into shared memory. Defenders should treat every read-write destination as part of the trust boundary. Useful protections include strict allowlists, destination validation, per-agent permissions, logging of external writes, and monitoring for repeated coordination patterns across agents.