Explainers
Questions security leaders ask about AI
Each explainer answers one question in its first two sentences, then shows the evidence. Every claim links to its source.
Q.What changed in AI regulation in September 2026?September 2026 tracker review: 14 entries against 1 in August 2026, with US States, US Federal and China milestones and what the instruments say.12 sources · 3 min read
Q.Does my AI system fall under the EU AI Act's Annex III high-risk list?A walkthrough of the EU AI Act's eight Annex III high-risk areas, quoted from the Act itself, plus the narrow exceptions that can take a matching system back out.4 sources · 6 min read
Q.What does ETSI TS 104 223 require of organizations that develop, deploy or operate AI systems, and how does it relate to EU AI Act standardisation?The two sources name ETSI's baseline AI security standard but list no requirements, and they do not link it to EU AI Act standardisation.2 sources · 5 min read
Q.What EU AI Act deadlines are already in force, and which are still ahead?Prohibited practices and GPAI rules are already in force. The Digital Omnibus moved Annex III high-risk rules to December 2027 and embedded products to August 2028.5 sources · 3 min read
Q.How do CISOs use the NIST AI Risk Management Framework?How security leaders apply the voluntary NIST AI RMF 1.0 and its four functions, Govern, Map, Measure and Manage, plus the Generative AI Profile, NIST AI 600-1.7 sources · 4 min read
Q.How do the EU AI Act and GDPR interact for security teams?A plain explainer on how EU data protection authorities, the AI Office and the AI Act divide oversight of personal data in AI systems.4 sources · 4 min read
Q.How do you secure AI agents?Secure AI agents with narrow tool access, short-lived credentials, human sign-off for high-impact actions and designs that expect prompt injection.8 sources · 3 min read
Q.How should a security leader set up an AI governance program?First steps for an AI governance program: named ownership, an AI inventory, risk tiers and data controls, drawn from NIST AI RMF, ISO/IEC 42001, OMB, CISA and NYDFS.8 sources · 4 min read
Q.How should companies defend against deepfake fraud?Deepfake fraud is beaten by process, not perception: verify requests out of band, protect payment approvals and train staff, as FBI, FinCEN and NSA guidance advises.7 sources · 3 min read
Q.How do NIST AI RMF, ISO/IEC 42001 and the EU AI Act compare?NIST's AI RMF is voluntary US guidance, ISO/IEC 42001 is a certifiable international standard, and the EU AI Act is binding law. What each requires, and how they map.6 sources · 6 min read
Q.What attack types and lifecycle stages does NIST's AI 100-2 E2025 adversarial machine learning taxonomy cover for predictive and generative AI/LLM systems, and how can a security team use it to structure threat modeling and mitigations?What attack types and lifecycle stages NIST AI 100-2 E2025 covers for predictive and generative AI, and how security teams can use its structure.2 sources · 5 min read
Q.What do the joint CISA, NSA, and partner principles for integrating AI into operational technology require, and how should a security leader apply them to OT and critical infrastructure environments?What the December 3, 2025 joint guide asks of critical infrastructure OT owners, and how to map its four principles onto your own program.2 sources · 5 min read
Q.What AI rules apply to US companies after Executive Order 14110 was revoked?EO 14110 was revoked in January 2025. Federal policy now favors light-touch AI rules, while Colorado, California, Texas, New York City and NYDFS impose real obligations.12 sources · 4 min read
Q.What are the EU AI Act penalties and fines?A breakdown of Article 99's three fine tiers, periodic penalties, enforcement split and the SME carve-out under the EU AI Act.5 sources · 5 min read
Q.What does the EU AI Act require of a CISO?The EU AI Act makes high-risk AI resist poisoning and evasion, sets 15-day incident reporting and log retention, and moves Annex III rules to 2 December 2027.6 sources · 4 min read
Q.What does the NYDFS AI cybersecurity guidance require of covered entities?How NYDFS's October 2024 letter and May 2026 frontier-AI advisory interpret Part 500 for AI-related cyber risk.5 sources · 5 min read
Q.What is AI data security and how do you protect training data?AI data security explained: the risks to training data and the best practices CISA, NSA, FBI and allied agencies recommend to secure it.5 sources · 3 min read
Q.What is ISO/IEC 42001 and should security teams pursue it?ISO/IEC 42001:2023 sets requirements for an AI management system. Certification is voluntary and done by independent bodies; here is how security teams can weigh it.7 sources · 3 min read
Q.What is prompt injection and how do you defend against it?Prompt injection is input that overrides an LLM's intended behaviour. OWASP ranks it LLM01:2025, and the UK NCSC says it may never be fully mitigated like SQL injection.5 sources · 3 min read
Q.What is shadow AI and how should CISOs govern employee use of generative AI?Shadow AI is staff use of AI tools outside approved systems. The UK NCSC advises cutting the risk with approved tools, open culture and data rules rather than bans alone.5 sources · 3 min read
Q.What is the Model Context Protocol (MCP) and what are its security risks?An explainer on MCP, the open protocol connecting AI agents to tools and data, and the access-control, token and injection risks it introduces.4 sources · 4 min read
Q.What is the OWASP Top 10 for LLM Applications?The OWASP Top 10 for LLM Applications ranks the ten most critical risks in apps built on large language models. The 2025 edition was published November 17, 2024.13 sources · 3 min read