AI Security for Cybersecurity Experts

AI systems fail differently from conventional software. Models are probabilistic, data is an attack surface, and behaviour can degrade without any attacker involved. Frameworks such as ISO 27001 were built for deterministic systems, and most security practitioners lack a method for fitting AI risks into them. This course bridges that gap, building on ENISA material and the FAICP methodology.

What you will learn

• Explain why AI systems need their own security, and describe the main threats and attack vectors against data, models and the application stack
• Apply a probabilistic security mindset, including security properties beyond the CIA triad and non-adversarial failures such as drift, and map it to deterministic frameworks like ISO 27001
• Classify AI-specific threats to models and data using a taxonomy aligned with ENISA terminology
• Threat model an AI/ML system to identify attack surfaces, adversaries and mitigations
• Identify risks and controls at each stage of the ML lifecycle, from data collection to deployment and monitoring
• Evaluate AI security use cases, select models based on security trade-offs, and use ENISA FAICP to translate requirements into controls compatible with existing standards

Who it is for

Security and risk professionals who already work with established cybersecurity standards and need to extend their practice to AI and ML systems: security architects, ISMS and compliance managers, risk officers, penetration testers and security engineers. No machine learning background is required; the course introduces the necessary AI concepts from a security perspective.