Before you can red-team an AI system or govern one, you need a map of what can go wrong. AI systems fail in ways ordinary software does not: they can be tricked by crafted input, leak the data they were trained on, be manipulated through poisoned data, or simply produce confident, harmful answers.
This module gives you that map from three angles. The OWASP Top 10 for LLM Applications ranks the most common weaknesses in apps built on language models, such as prompt injection and sensitive information disclosure. MITRE ATLAS catalogues the tactics and techniques real adversaries use against AI systems, in the style of ATT&CK. The NIST AI Risk Management Framework then shows how organisations govern, map, measure, and manage those risks.
Watch the lessons in order, then continue on to Threat Modeling AI Systems, where you will apply these frameworks to a real system.
A big-picture look at how AI helps attackers and defenders alike, including convincing phishing, over-reliance on AI output, and why security controls still matter.
A conference talk on the most critical vulnerabilities in LLM applications, with practical guidance on defending against them.
How MITRE ATLAS extends the ATT&CK approach to map the tactics and techniques used against AI systems.
An introduction to the NIST AI Risk Management Framework: the attributes of trustworthy AI and the govern, map, measure, and manage functions.