AI Security Wiki GitHub

Home / EU AI Compliance / Guide for Implementors: Building an EU AI Governance Programme

Guide for Implementors: Building an EU AI Governance Programme

Verification required. EU claims on this page were compiled on 7 October 2026 from secondary sources and from memory of the legal text. The Official Journal text was not read, and each claim is tagged [Reported], [Recalled], [Conflict] or [Unverified] as explained on the section home page. Check the primary text on EUR-Lex before relying on any of them. This is not legal advice.

For programme leads, GRC teams, CISOs, DPOs and virtual CISOs. EU claims carry the tags used in the EU AI Regulatory Hub.

1. Target outcome in 90 days

An inventory with role and tier, a literacy programme, a policy, an impact assessment procedure, a supplier addendum, an incident workflow, named owners, and a register of dates and authorities.

2. What applies when

A planning view. All dates are [Reported].

When What
Now Prohibitions, AI literacy, GPAI duties, transparency (check the date nuance)
Dec 2026 Marking grace ends for existing generative systems; new prohibition on intimate-imagery generation
Dec 2027 High-risk Annex III
Aug 2028 High-risk Annex I products

Do not wait for the late dates. Inventory, classification and documentation take months.

3. 90-day plan

Days Workstream Output Template
0 to 15 Discover Inventory v1, including AI embedded in vendor products T09
0 to 15 Mandate Sponsor, steering group, charter none
0 to 30 Prohibited screen All use cases checked against Art. 5 T01
15 to 45 Classify Role and tier per system T01
15 to 60 Literacy Role-based training and records T07
30 to 60 Assess DPIA and FRIA procedure; first assessments T02
30 to 60 Suppliers Addendum issued to AI suppliers T03
45 to 75 Data Transfer record per system T04
45 to 75 Incident Multi-regime workflow; tabletop exercise T05
60 to 90 Transparency Notices and marking review T06
60 to 90 Documentation Technical file skeleton for each high-risk system T10
75 to 90 Assure Internal audit pass; board report T08, T11

4. RACI (starting point)

Activity Board AI gov lead CISO DPO Legal Business owner Engineering
Policy A R C C C I I
Inventory I A C C I R R
Classification I A C C R R C
DPIA and FRIA I A C R C R C
Supplier due diligence I C R C C A I
Technical documentation I A C C I C R
Incident notification I C A R R C R
Literacy I A C C C R I
Authority engagement A R C C R I I

5. Decisions you must make early

  1. The spine framework. Use the wiki controls mapped to the AI Act, with ISO/IEC 42001 as a certifiable bridge. ISO/IEC 42001 is not a harmonised standard [Recalled].
  2. A provider or deployer posture for each system. Avoid becoming a provider by accident.
  3. A definition of "high-impact" that works across GDPR Art. 22 and AI Act Annex III.
  4. Approved hosting regions and a transfer position.
  5. A joint DPIA and FRIA procedure (T02).
  6. Who may sign technical documentation and declarations.

6. Maturity steps

Level Looks like Evidence
1 Ad hoc none
2 Inventory, policy, classification T01, T09
3 Assessments, supplier controls, literacy T02, T03, T07
4 Technical files, internal audit, authority log T10, T08, T11
5 Continuous monitoring; certification where useful ISO/IEC 42001 certificate, if pursued

7. Metrics

  • Percent of systems with a role and a tier
  • High-risk systems with a complete technical file
  • Staff who have completed role-based literacy training
  • Suppliers with a signed addendum
  • Time to classify an incident and send notices
  • Systems with a tested kill switch
  • Open audit findings that are overdue

8. Overlaps to integrate

NIS2, DORA and the CRA share risk management, reporting and supply-chain themes with the AI Act [Reported]. Build one incident process (T05) and one supplier process (T03) rather than four of each. One source claims this saves 30 to 40 percent of the effort. That figure rests on a single source and is not established [Unverified].

9. Pitfalls

  • Waiting for the late dates.
  • Classifying by marketing label.
  • Ignoring AI embedded in vendor products.
  • Running literacy training once and never again.
  • Writing documentation after launch.
  • Treating the voluntary marking code or the draft guidelines as final law.

10. Stop triggers

Stop or pause a system on a failed threshold, an unlawful data location, an unresolved serious incident, an authority instruction, or the discovery of a prohibited practice. Link each trigger to kill-switch ownership (AI-CTRL-025).