The AI Security Framework isn’t a single document — it’s a stack of frameworks that together define how organisations secure AI systems. Based on authoritative sources, the recognised AI security landscape is built from five major frameworks, each covering a different layer of AI risk, governance, and adversary behaviour.
The most important takeaway: AI security = governance + risk management + engineering controls + threat intelligence + legal obligations. No single framework covers all of this.
What it is: A risk‑management process for AI systems. Purpose: Helps organisations govern, map, measure, and manage AI risks. Status: Voluntary, not certifiable. Use it for: Building a defensible AI risk programme.
What it is: The world’s first certifiable AI governance standard. Purpose: Defines how organisations should manage AI responsibly and securely. Status: Certifiable via audit. Use it for: Demonstrating compliance and governance maturity to regulators and customers.
What it is: A threat‑intelligence catalogue of real attacks on AI systems. Purpose: Shows how attackers target AI models, data pipelines, and inference systems. Status: Non‑certifiable knowledge base. Use it for: Threat modelling and detection engineering for AI systems.
What it is: Engineering guidance for securing AI and LLM applications. Purpose: Provides practical controls, threat maps, and secure development patterns. Status: Community standard. Use it for: Securing AI applications, preventing prompt injection, data leakage, model abuse, etc.
What it is: Legally binding regulation governing AI systems in the EU. Purpose: Defines prohibited practices, high‑risk requirements, transparency duties, and penalties. Status: Law — fines up to €35m or 7% of global turnover. Use it for: Understanding legal obligations and compliance requirements.