Adversarial AI and threat hunting

Security tools now rely on machine learning to spot attacks. Learn why that makes the models themselves a target, and how defenders map those attacks with MITRE ATLAS and NIST’s taxonomy.

Every classifier has a line between “malicious” and “benign”. Understand why repeated black-box queries can reveal where that line sits, and the controls that make probing slow, noisy and visible.

Attackers can fool a model at decision time (evasion) or corrupt what it learns (poisoning). Learn how each works in principle and the defensive practices that make models harder to fool.

A detection model that misses an attack produces no alert, no log line and no ticket. Learn why that silence is the real danger, and how to design security that still sees an attacker after a miss.

The answer to attacks on AI is not a better single model but several independent checks. Learn how behavioural detection, input validation and human review combine so no single failure is fatal.

Threat hunting assumes an attacker may already be inside and goes looking. Learn how threat intelligence and MITRE ATT&CK turn that search into testable hypotheses instead of random browsing through logs.

Put the course together in a discussion-based exercise: a fictional company, an attack that slipped past its ML detection, a series of injects, and the questions a defending team should answer, with model answers.