Multi-session misuse is harmful or policy-violating behavior that unfolds across more than one interaction, account, or time window. Unlike a single abusive prompt, the activity is distributed: an attacker may probe a system in one session, refine the approach in another, and execute the harmful step later. This pattern matters in cyber security because simple request-level filters can miss slow, coordinated abuse.
In real systems, multi-session misuse can look like repeated reconnaissance, incremental phishing or social-engineering prompts, staged exfiltration attempts, or an agent being guided to bypass safeguards over time. Defenders look for shared identifiers, unusual timing, repeated themes, and correlated behavior across sessions rather than relying only on content inspection. This is especially important when privacy controls reduce access to raw prompts, forcing security teams to detect abuse through metadata, telemetry, and policy-aware correlation.