AI Security Monitoring and Incident Response Course
IDCYSE8014

A focused practitioner course on the two things traditional IT monitoring and incident response handle poorly for AI systems: recognizing what a compromise looks like, and knowing what to do about it. The organizing idea is that an AI system inherits every traditional attack surface and then adds more, tied to data, models and inference behavior. An attacker can manipulate what a model learns from, or what it sees at run time, without ever reaching the infrastructure. Monitoring has to shift accordingly, from "is the system running" to "is the system behaving as intended, on data it should be seeing, for the purposes it was approved for."

The most useful thing participants take away is the ability to tell model drift from an actual attack. A signals table maps observed patterns to likely causes — a sustained spike in near-identical unusual inputs suggests adversarial probing, a slow accuracy decline suggests drift, an unexplained query spike from a single credential suggests an extraction attempt — and it pairs with a first-hour triage sequence. The course then works on judgment: separating anomalies from incidents, preserving the evidence responders most often lose (the model version in use, the surrounding prompt and response logs, training data lineage, and the guardrail configuration in effect), and classifying severity by impact rather than technical complexity. That last point changes behavior: an incorrect benefits determination caused by a data integrity failure is high severity even though nothing technically broke. The closing module applies the four-phase incident response lifecycle with AI-specific containment — model rollback to a validated version, disabling automated decision-making in favor of human review, quarantining a compromised training data source, and closing a confirmed prompt injection vector. It follows directly from AI Security & Governance for Government Course, making a clean two-course sequence: governance first, then operations.