AI Security
This course examines the security of AI systems, covering the principles, practices and policies involved in building and deploying machine-learning-based systems that remain trustworthy under attack. It looks at how AI components change the attack surface of a system, and at what technical measures and governance are needed in response.
The course is examined through a group written report followed by an individual oral exam.
About the course
The course is part of the Security specialisation of the MSc in Computer Science at the IT University of Copenhagen (7.5 ECTS, taught in autumn). It is also open to guest students, exchange students and single-subject students.
Learning objectives
After taking this course, students will be able to:
- identify security and privacy threats specific to machine-learning systems
- assess the robustness of AI-based systems against attack
- apply technical and organisational measures to secure AI systems
- relate AI security practice to relevant policy and regulation
Topics
- Threat models for machine-learning systems
- Attacks on models, training data and deployed systems
- Defences and secure development practice
- Policy, regulation and governance of AI security