A working library, not a link dump.
Every AI governance conversation eventually needs the same handful of primary sources — the actual regulatory text, the actual standard, the actual threat model — not someone's summary of it. This is that library: regulation, standards, thought leadership, study material, and reference tools, each with a short note on why it's here and who it's actually for. Reviewed quarterly, so a stale link is a bug, not a feature.
The essentials, if you read nothing else.
One pass across all five categories — the documents and voices that come up in almost every governance conversation this practice has.
EU AI Act — Regulation (EU) 2024/1689(opens in a new tab)
The official, full legal text of the EU AI Act, entered into force 1 August 2024, staggered into application through August 2027.
Take: This is the primary source — cite this, not a law firm's summary of it, when precision matters. It's dense; pair it with the Service Desk timeline below before attempting to read it cover to cover.
- Compliance
- Security
A pro-innovation approach to AI regulation(opens in a new tab)
The UK's foundational statement of intent: five cross-sectoral principles, existing-regulator-led enforcement, no dedicated AI Act.
Take: If you only read one document to understand why the UK's approach looks nothing like the EU's, this is it. Still the reference point even as sector regulators (FCA, ICO) build on it.
- Board & C-suite
- Compliance
Guidance on AI and data protection(opens in a new tab)
The ICO's detailed guidance on applying UK GDPR to AI systems, plus a practical AI and data protection risk toolkit.
Take: Underused. Most AI governance conversations jump straight to the EU AI Act and skip the fact that, for UK organisations, data protection law already bites hard on most AI use cases today.
- Compliance
- Security
The Mills Review — AI and the future of retail financial services(opens in a new tab)
The FCA's landmark review of AI adoption and the autonomy spectrum in UK retail financial services.
Take: Essential if you're anywhere near UK financial services — this is the document the site's own Thought Experiment investigation is built on. Read it before you read my commentary on it, not after.
- Board & C-suite
- Compliance
ISO/IEC 42001:2023 — AI management systems(opens in a new tab)
The world's first certifiable AI management system standard (AIMS) — Annex SL structure, 38 Annex A controls.
Take: The backbone of this practice's governance advisory work. If you're building an AIMS from scratch, start here, not with a framework mapping exercise.
- Compliance
- Board & C-suite
NIST AI Risk Management Framework (AI RMF 1.0)(opens in a new tab)
Voluntary US framework structured around four functions: Govern, Map, Measure, Manage.
Take: More approachable than ISO 42001 for a first read, and the Govern/Map/Measure/Manage structure is a genuinely useful mental model even if you end up certifying against 42001 instead.
- Board & C-suite
- Compliance
- Newcomers
OWASP Top 10 for LLM Applications(opens in a new tab)
The standard awareness document for LLM application security risks, now part of the broader OWASP GenAI Security Project.
Take: If a vendor pitches you an "AI security" product and can't explain how it maps to this list, that's informative in itself.
- Security
- Practitioners
MITRE ATLAS(opens in a new tab)
A living knowledge base of adversary tactics and techniques targeting AI systems, modelled on MITRE ATT&CK.
Take: This is the one I actually use for threat modelling, not just citing. If your security team already thinks in ATT&CK, ATLAS is a five-minute onboarding, not a new discipline.
- Security
- Practitioners
Stanford HAI — AI Index Report(opens in a new tab)
An independent, annual, data-driven snapshot of AI progress, investment, and — increasingly — governance readiness.
Take: The single best "what actually happened this year" document in the field. The governance/regulation chapters are usually the most sobering part of the report, not the capability charts.
- Board & C-suite
- Newcomers
Ada Lovelace Institute(opens in a new tab)
Independent UK research institute producing policy-facing research on AI governance and regulation.
Take: The UK's best-calibrated policy voice — less breathless than most Silicon Valley commentary, more rigorous than most law-firm client alerts. Their EU AI Act analysis shaped a lot of UK thinking.
- Board & C-suite
- Compliance
Import AI(opens in a new tab)
Weekly newsletter tracking frontier AI research and its policy implications, from someone who has sat on both the lab and policy sides.
Take: The weekly discipline of "what actually got published, and why it matters" is more useful than most monthly think-pieces. The short fiction he includes is a genuinely good way to pressure-test governance assumptions against plausible futures.
- Practitioners
- Security
AI Snake Oil(opens in a new tab)
A rigorous, skeptical field guide to what AI can and can't actually do, and how to tell the difference.
Take: The single best gift for a board member who's been on the receiving end of one too many AI vendor pitches. Required reading before signing off on any AI procurement.
- Board & C-suite
- Newcomers
AIGP — Artificial Intelligence Governance Professional(opens in a new tab)
The IAPP's flagship AI governance certification — no formal prerequisites, 100-question exam.
Take: The closest thing the field has to a recognised baseline credential right now. If you're building a governance career and can only get one AI-specific certification, this is the safest first choice.
- Compliance
- Newcomers
AI For Everyone(opens in a new tab)
A short, non-technical course for understanding what AI can and can't do and spotting opportunities to apply it.
Take: Still the best answer to "where should a non-technical exec start?" years after it launched — no coding, no jargon, correctly calibrated on hype.
- Board & C-suite
- Newcomers
AI Incident Database(opens in a new tab)
A searchable, crowdsourced database of over 1,200 real-world AI harm incidents.
Take: The single best tool for grounding a risk assessment in what has actually gone wrong, rather than hypothetical failure modes. Search it before you write a risk register from a blank page.
- Security
- Compliance
- Practitioners
UK AI Security Institute (AISI)(opens in a new tab)
The UK government's technical AI safety and security research body (formerly the AI Safety Institute).
Take: Their blog is where the UK government's actual technical thinking on frontier AI risk shows up first, well before it filters into policy documents.
- Security
- Practitioners
The rules you're actually bound by.
Binding rules and official government/regulator positions — the things you cite in a compliance file, not a blog post.
EU AI Act — Regulation (EU) 2024/1689(opens in a new tab)
The official, full legal text of the EU AI Act, entered into force 1 August 2024, staggered into application through August 2027.
Take: This is the primary source — cite this, not a law firm's summary of it, when precision matters. It's dense; pair it with the Service Desk timeline below before attempting to read it cover to cover.
- Compliance
- Security
EU AI Act Service Desk — Implementation Timeline(opens in a new tab)
The Commission's own tracker of which obligations are live now versus deferred.
Take: The single most useful page for answering "wait, is this bit in force yet?" — better than any third-party summary because it's the regulator's own answer, updated as the Digital Omnibus amendments land.
- Compliance
- Board & C-suite
A pro-innovation approach to AI regulation(opens in a new tab)
The UK's foundational statement of intent: five cross-sectoral principles, existing-regulator-led enforcement, no dedicated AI Act.
Take: If you only read one document to understand why the UK's approach looks nothing like the EU's, this is it. Still the reference point even as sector regulators (FCA, ICO) build on it.
- Board & C-suite
- Compliance
AI Playbook for the UK Government(opens in a new tab)
Practical, cross-technology guidance for public-sector AI adoption — not just generative AI.
Take: Written for civil servants, but genuinely useful as a template for how to write internal AI usage guidance that isn't either useless platitudes or unreadable legalese.
- Compliance
- Practitioners
Guidance on AI and data protection(opens in a new tab)
The ICO's detailed guidance on applying UK GDPR to AI systems, plus a practical AI and data protection risk toolkit.
Take: Underused. Most AI governance conversations jump straight to the EU AI Act and skip the fact that, for UK organisations, data protection law already bites hard on most AI use cases today.
- Compliance
- Security
The Mills Review — AI and the future of retail financial services(opens in a new tab)
The FCA's landmark review of AI adoption and the autonomy spectrum in UK retail financial services.
Take: Essential if you're anywhere near UK financial services — this is the document the site's own Thought Experiment investigation is built on. Read it before you read my commentary on it, not after.
- Board & C-suite
- Compliance
SS1/21 — Operational Resilience(opens in a new tab)
Supervisory expectations on impact tolerances, mapping, and scenario testing for important business services.
Take: Doesn't mention AI once, and that's exactly the point — most firms' AI governance gap is that they haven't mapped where AI already sits inside an "important business service" under this framework.
- Compliance
- Board & C-suite
SS2/21 — Outsourcing and Third-Party Risk Management(opens in a new tab)
Expectations for material outsourcing and third-party risk, including cloud and new technology adoption.
Take: If your AI risk is really "a vendor's model, not yours" — which it usually is — this is the framework that already tells you what due diligence looks like. Don't reinvent it for AI specifically.
- Compliance
- Board & C-suite
OECD AI Principles(opens in a new tab)
The first intergovernmental AI standard (2019, updated 2024) — five values-based principles, 47 adherent countries.
Take: Soft law, not binding, but it's the common ancestor most national frameworks (including the EU AI Act's risk-based approach) trace back to. Useful for explaining why everyone's framework rhymes.
- Board & C-suite
- Newcomers
The frameworks you implement against.
Voluntary technical and management-system standards — the ones you actually implement against.
ISO/IEC 42001:2023 — AI management systems(opens in a new tab)
The world's first certifiable AI management system standard (AIMS) — Annex SL structure, 38 Annex A controls.
Take: The backbone of this practice's governance advisory work. If you're building an AIMS from scratch, start here, not with a framework mapping exercise.
- Compliance
- Board & C-suite
ISO/IEC 23894:2023 — AI risk management guidance(opens in a new tab)
Guidance (not a certifiable standard) on managing AI-specific risk, built on ISO 31000 principles.
Take: The natural companion to 42001's Clause 6 risk-treatment requirement — this is where the "how" actually lives. Don't skip it just because it isn't certifiable.
- Compliance
- Practitioners
NIST AI Risk Management Framework (AI RMF 1.0)(opens in a new tab)
Voluntary US framework structured around four functions: Govern, Map, Measure, Manage.
Take: More approachable than ISO 42001 for a first read, and the Govern/Map/Measure/Manage structure is a genuinely useful mental model even if you end up certifying against 42001 instead.
- Board & C-suite
- Compliance
- Newcomers
NIST Generative AI Profile(opens in a new tab)
An AI RMF companion profile addressing risks specific to generative AI.
Take: Technical, but worth it if your risk register is still written as if "AI" only means classic ML. Generative and agentic systems have a different failure mode list.
- Security
- Practitioners
OWASP Top 10 for LLM Applications(opens in a new tab)
The standard awareness document for LLM application security risks, now part of the broader OWASP GenAI Security Project.
Take: If a vendor pitches you an "AI security" product and can't explain how it maps to this list, that's informative in itself.
- Security
- Practitioners
MITRE ATLAS(opens in a new tab)
A living knowledge base of adversary tactics and techniques targeting AI systems, modelled on MITRE ATT&CK.
Take: This is the one I actually use for threat modelling, not just citing. If your security team already thinks in ATT&CK, ATLAS is a five-minute onboarding, not a new discipline.
- Security
- Practitioners
Guidelines for Secure AI System Development(opens in a new tab)
"Secure by design" guidance across the AI system development lifecycle.
Take: Rare example of a joint UK/US government document that's actually well-written and short enough to hand to an engineering team without losing them.
- Security
- Practitioners
Who's worth reading regularly.
The voices worth reading regularly — chosen for rigour, not volume.
Stanford HAI — AI Index Report(opens in a new tab)
An independent, annual, data-driven snapshot of AI progress, investment, and — increasingly — governance readiness.
Take: The single best "what actually happened this year" document in the field. The governance/regulation chapters are usually the most sobering part of the report, not the capability charts.
- Board & C-suite
- Newcomers
Ada Lovelace Institute(opens in a new tab)
Independent UK research institute producing policy-facing research on AI governance and regulation.
Take: The UK's best-calibrated policy voice — less breathless than most Silicon Valley commentary, more rigorous than most law-firm client alerts. Their EU AI Act analysis shaped a lot of UK thinking.
- Board & C-suite
- Compliance
The Alan Turing Institute(opens in a new tab)
The UK's national institute for data science and AI.
Take: Academic in tone, but their public sector and safety research is directly usable — worth following even if you skip most of the pure-research output.
- Practitioners
- Newcomers
Import AI(opens in a new tab)
Weekly newsletter tracking frontier AI research and its policy implications, from someone who has sat on both the lab and policy sides.
Take: The weekly discipline of "what actually got published, and why it matters" is more useful than most monthly think-pieces. The short fiction he includes is a genuinely good way to pressure-test governance assumptions against plausible futures.
- Practitioners
- Security
Don't Worry About the Vase(opens in a new tab)
Weekly, exhaustively detailed AI policy and capability tracking.
Take: Not for the time-poor — this is the deep-end option. Read it when you need the full context behind a headline, not the headline itself.
- Practitioners
One Useful Thing(opens in a new tab)
Practical, research-grounded commentary on AI's implications for work, education, and organisational adoption.
Take: The best antidote to both AI hype and AI dismissal — Mollick actually uses the tools and reports what happens, which is rarer than it should be at executive level.
- Board & C-suite
- Newcomers
AI Snake Oil(opens in a new tab)
A rigorous, skeptical field guide to what AI can and can't actually do, and how to tell the difference.
Take: The single best gift for a board member who's been on the receiving end of one too many AI vendor pitches. Required reading before signing off on any AI procurement.
- Board & C-suite
- Newcomers
AI Governance Alliance — Briefing Paper Series(opens in a new tab)
Multi-stakeholder briefing papers on responsible AI governance from a 250+ member cross-industry alliance.
Take: Uneven — some papers are genuinely sharp, others read like a committee wrote them, because a committee did. Worth scanning titles rather than subscribing to everything.
- Board & C-suite
Where to build the credential.
Where to actually build the credential, not just the opinion.
AIGP — Artificial Intelligence Governance Professional(opens in a new tab)
The IAPP's flagship AI governance certification — no formal prerequisites, 100-question exam.
Take: The closest thing the field has to a recognised baseline credential right now. If you're building a governance career and can only get one AI-specific certification, this is the safest first choice.
- Compliance
- Newcomers
AAIR — Advanced in AI Risk(opens in a new tab)
Advanced credential for experienced IT risk professionals managing AI-specific risk, requiring a prerequisite certification (CRISC, CISA, CISM, etc.).
Take: Not an entry-level credential, and shouldn't be treated as one — the prerequisite gate is doing real work here. Worth it once you already have the risk-management foundation and need the AI-specific layer.
- Compliance
- Practitioners
AAISM — Advanced in AI Security Management(opens in a new tab)
ISACA's AI-centric security management credential, building on CISM/CISSP foundations.
Take: The security-side sibling to AAIR. If your background is CISSP rather than CRISC, this is the more natural next step.
- Security
- Practitioners
ISO/IEC 42001 Senior Lead Implementer(opens in a new tab)
Accredited training and certification for implementing and leading ISO/IEC 42001 AIMS build-outs.
Take: The practical, implementation-focused counterpart to the AIGP's broader governance framing. Better suited to someone who's going to actually build the management system, not just advise on one.
- Compliance
- Practitioners
AI For Everyone(opens in a new tab)
A short, non-technical course for understanding what AI can and can't do and spotting opportunities to apply it.
Take: Still the best answer to "where should a non-technical exec start?" years after it launched — no coding, no jargon, correctly calibrated on hype.
- Board & C-suite
- Newcomers
Generative AI for Everyone(opens in a new tab)
The generative-AI-specific follow-up, covering prompting, workflow redesign, and a business-strategy framework.
Take: Do "AI For Everyone" first if you haven't — this assumes you already have the baseline mental model and moves straight to application.
- Board & C-suite
- Newcomers
Books worth the time — no single canonical link
- The Alignment Problem — Brian Christian. The best plain-language explanation of why "make AI do what we want" is harder than it sounds.
- Human Compatible — Stuart Russell. Denser, written by one of the field's most credible technical voices; good after The Alignment Problem, not instead of it.
- Weapons of Math Destruction — Cathy O'Neil. Pre-dates the generative AI wave but is still the sharpest account of how algorithmic systems cause real-world harm at scale — the governance failure modes it describes haven't gone away, they've just got a new interface.
The rest of the desk.
Tools, databases, and institutions that don't fit neatly into the other four categories but earn a place on the desk anyway.
AI Incident Database(opens in a new tab)
A searchable, crowdsourced database of over 1,200 real-world AI harm incidents.
Take: The single best tool for grounding a risk assessment in what has actually gone wrong, rather than hypothetical failure modes. Search it before you write a risk register from a blank page.
- Security
- Compliance
- Practitioners
UK AI Security Institute (AISI)(opens in a new tab)
The UK government's technical AI safety and security research body (formerly the AI Safety Institute).
Take: Their blog is where the UK government's actual technical thinking on frontier AI risk shows up first, well before it filters into policy documents.
- Security
- Practitioners
MITRE ATT&CK(opens in a new tab)
The foundational adversary tactics/techniques knowledge base that MITRE ATLAS extends into AI systems.
Take: Include this even though it predates AI, because most security teams already have ATT&CK muscle memory — ATLAS only clicks quickly if you already understand the parent framework.
- Security
Partnership on AI(opens in a new tab)
A multi-stakeholder nonprofit working on responsible AI norms and practice, including the AI Incident Database.
Take: Membership skews toward the labs themselves, so read their outputs knowing whose interests are in the room — still useful, just not neutral in the way a regulator is.
- Board & C-suite
- Practitioners
arXiv — cs.AI (recent)(opens in a new tab)
The open-access preprint firehose for AI research, unfiltered and unreviewed.
Take: Not for casual browsing — this is for when you need to check whether a specific technique or attack has published research behind it. Import AI does the filtering most people actually want.
- Practitioners
- Security
Spot something stale? Suggest an addition.
This library is reviewed quarterly. If a link has moved, or you think something essential is missing, tell me.
Last reviewed: 2026-07-22