AI fundamentals and applications across the CC domains

AI is broader than chatbots

The selected NIST introduction treats an AI system in terms of producing outputs such as predictions, recommendations or decisions for objectives. AI includes multiple techniques and use cases. Machine learning, generative models, large language models and agentic applications are related terms, not synonyms. Generative AI can work with images, audio or other forms as well as language; not every generative model is an LLM.

Context changes the risk

Data, human choices, organizational use and surrounding systems affect benefits and harms. Treat AI as part of a sociotechnical system: a plausible output is not verified evidence, and a model’s behaviour cannot be assessed without its use context. An application described as an agent may have different degrees of autonomy and tools; no single LLM-based design is universal.

Apply the foundations already learned

Editorial application: Identify information flows, permissions, dependencies and the consequences of incorrect or inappropriate output. Distinguish a model capability claim from a tested control or business outcome. This section introduces the topic only; it does not provide a complete AI-governance framework or establish compliance with AI legislation.

Use the official CC AI context with the numbered outline

ISC2’s September 2026 guidance applies AI to every CC domain: principles and governance; resilience and awareness; non-human identities; network/cloud responsibilities; and operations, response and testing. Read the CC-specific guidance alongside the numbered outline. These applications do not add a sixth domain, change the published weights or establish a separate AI exam prerequisite.

Distinguish data poisoning from model poisoning

NIST describes poisoning as adversarial interference with training. Manipulating training data is one route; direct manipulation of a model or its updates is another. Such attacks can damage integrity or availability. Protect the relevant data, model and process in context rather than assuming one file-integrity check prevents every poisoning attack. Detailed attack methods and mitigation recipes are outside this introductory assignment.

Monitor performance without assuming an attack

An AI system can produce unsuitable results when deployed outside the conditions in which it was evaluated. Relevant data, conditions or behaviour may change over time; drift is a reason to monitor and investigate, not an automatic finding of compromise. NIST treats validity/reliability as contextual and calls for monitoring production functionality. Its maintenance discussion explicitly notes data, model and concept drift. Full drift diagnosis and model maintenance remain beyond this introduction.

Balance the relevant trustworthiness properties

NIST’s AI RMF includes validity/reliability, safety, security/resilience, accountability/transparency, explainability, privacy and management of harmful bias. The use context and trade-offs matter; addressing one property alone does not establish trustworthiness. For a beginner case, explain which property matters, what evidence is available and what decision needs qualified review. This is background, not a complete AI RMF, legal-compliance or ISO/IEC 42001 course.

Connect the terminology.