Agentic abuse is the misuse of autonomous or semi-autonomous AI systems to perform repeated actions against a target platform. Instead of a one-time prompt or isolated request, the attacker uses the agent’s ability to browse, decide, retry, and adapt over time. That makes the abuse look more like ordinary user activity and less like a single obvious intrusion.
This matters in cybersecurity because agents can amplify scale, persistence, and timing. They may post content, submit forms, probe controls, or coordinate tasks while staying within nominal permissions. Defenders look for this pattern with strong identity checks, rate limiting, action logging, and anomaly detection that tracks sequences of behavior, not just single events. In practice, agentic abuse is a governance problem as much as a model problem: if an AI tool can act on a site, it needs the same restrictions and monitoring applied to other high-value interfaces.