AI Safety, Reliability, and Security

About the course

This course treats AI safety as an engineering and research problem: from the technical aspects of adversarial robustness and jailbreak attacks to the systemic questions of alignment and governance. It covers both defensive techniques (red teaming, constitutional AI, interpretability) and regulatory frameworks (EU AI Act, NIST AI RMF). The course prepares you for responsible deployment of AI systems and is relevant to any career track.

What you'll learn

✓Run a red team assessment of an LLM system and document the vulnerabilities found

✓Apply differential privacy to an ML pipeline with quantitative guarantees

✓Assess an AI system against EU AI Act and NIST AI RMF requirements

✓Implement a baseline defense against adversarial attacks (adversarial training or certified defense)

Key topics

•Adversarial examples: attacks (FGSM, PGD, AutoAttack) and defenses (adversarial training, certified defenses)

•Prompt injection and LLM jailbreaking: mechanisms and mitigation

•Alignment: RLHF, Constitutional AI, scalable oversight

•Interpretability and mechanistic interpretability: circuits, features

•Red teaming AI systems: methodology and tools

•Privacy in ML: differential privacy (DP), federated learning, membership inference

•Regulatory frameworks: EU AI Act, NIST AI RMF, ISO/IEC 42001

•Supply chain attacks on ML: data poisoning, model poisoning

•AI governance, model cards, and system cards

This description was generated automatically and has not yet been reviewed by an instructor — it's a draft for discussion.

About the course

What you'll learn

Key topics

Feedback