How to Build a Compliance Map for EU and UK AI Rules
If you sell AI products on both sides of the Channel, learning how to build a compliance map is the most useful thing you can do this quarter. At the November 20 summit in Brussels, the EU and UK agreed to launch a Digital Partnership early next year covering AI security and safety, digital wallets, trust services and online safety, while each side keeps its own rules.
That last part is the key. The joint statement talks about cooperation “while safeguarding our regulatory autonomy.” So you will still face two regimes, but more chances to reuse the same evidence for both. A compliance map is how you capture those chances instead of doing everything twice. Here is how to build one.
What You Need Before You Start
- A list of your AI products and features: Including internal tools, third-party models you embed and any general-purpose AI you fine-tune.
- Where you sell and who uses it: EU member states, UK, or both, and whether users are consumers, businesses or public bodies.
- Your current compliance documents: GDPR and UK GDPR records of processing, DPIAs, security policies, any ISO 27001 or ISO/IEC 42001 work.
- An owner: One person (legal, compliance or a senior engineer) who keeps the map current.
A spreadsheet is fine to start. Some teams move to a GRC tool later, but don’t let tool shopping delay the work.
Step 1: Inventory and Classify Every AI System
List each AI system with its purpose, users, data inputs, outputs and the model behind it. Then classify it under the EU AI Act’s risk tiers: prohibited, high-risk (Annex III uses like hiring, credit scoring, education or critical infrastructure), limited-risk with transparency duties (chatbots, deepfakes, emotion recognition), or minimal risk. The EU AI Act Explorer is a handy way to check the text article by article.
Note the timing too. Prohibited practices and AI literacy duties already apply, general-purpose AI obligations have applied since August 2025, and the Digital Omnibus pushed key high-risk obligations back to 2027, as Gibson Dunn explains. A delay is breathing room, not a reason to stop.
Step 2: List the Obligations on Each Side
Now build two columns. On the EU side, the usual suspects are the AI Act, GDPR, the Digital Services Act if you host user content, the Data Act if you make connected products, and NIS2 if you are in a covered sector.
On the UK side, there is no single AI law. The UK relies on existing regulators applying shared principles (safety, transparency, fairness, accountability, contestability). In practice that means UK GDPR and the Data (Use and Access) Act 2025, guidance from the ICO on AI and data protection, the Online Safety Act for platforms, and sector rules from bodies like the FCA, CMA and Ofcom.
Write each obligation as a concrete requirement, for example “explain automated decisions to the affected person” rather than “transparency”. Concrete wording makes the next step possible.
Step 3: Map Obligations to Shared Controls
This is where the savings come from. For every obligation, ask: what control or document satisfies it? You will find heavy overlap. A single, well-written risk assessment can feed an AI Act risk management file, a GDPR DPIA and a UK GDPR DPIA. One set of model test results can support AI Act accuracy and robustness requirements and a UK regulator’s safety expectations.
Build a matrix with obligations as rows and controls as columns. Typical shared controls include:
- AI system inventory and risk classification
- Data governance and training data documentation
- Model evaluation, red-teaming and bias testing
- Human oversight procedures
- Incident logging and response
- User-facing transparency notices
- Vendor and model supply-chain due diligence
Mark each cell as “covers EU”, “covers UK” or “covers both”. The cells marked “both” are where you stop duplicating effort.
Step 4: Flag the Gaps and Divergences
Some obligations won’t overlap, and that’s fine. The EU AI Act’s conformity assessment, CE marking and EU database registration for high-risk systems have no direct UK equivalent. The UK Online Safety Act’s specific duties differ from the DSA’s. Data transfer mechanics differ too.
Highlight these in a separate list with an owner and a deadline for each. Be honest about what is market-specific. That list becomes your real compliance roadmap, and it’s usually much shorter than teams expect once the shared controls are mapped.
Step 5: Build One Evidence Pack With Two Front Pages
Store all your evidence (test reports, DPIAs, policies, training logs, incident records) in one place with version control. Then create two short cover documents, one for EU buyers and regulators and one for UK, each pointing to the relevant evidence.
When an enterprise customer in Frankfurt or Manchester sends a procurement questionnaire, you answer from the same pack. For AI safety testing specifically, keep an eye on the work of the UK’s AI Security Institute and the EU AI Office, since joint evaluation methods are exactly what the Digital Partnership could produce.
Step 6: Review Quarterly and Track the Partnership
A compliance map goes stale quickly. Review it every quarter and whenever you launch a new AI feature, change a model or enter a new market. Add a standing agenda item for the Digital Partnership: when working groups publish anything on AI testing, digital wallets or trust services, check which cells in your matrix it affects.
If the two sides do align safety testing next year, teams with a map will update a few cells. Teams without one will start from scratch.
Common Mistakes to Avoid
Assuming the partnership merges the rules. It doesn’t. GDPR and UK GDPR stay separate, and the AI Act applies only in the EU. Don’t cancel any compliance projects because of a summit statement.
Mapping laws instead of systems. A map that lists regulations without tying them to specific AI systems is just a reading list. Always start from what you actually build and run.
Ignoring third-party models. If you build on a general-purpose model, part of your compliance depends on your provider’s documentation. Ask for it early, and record the model and version in your inventory.
Treating it as a legal-only job. Engineers produce most of the evidence: logs, test results, data lineage. Bring them in from day one, or the map will describe controls that don’t exist.
Key Takeaways
- Two regimes remain: The EU-UK Digital Partnership encourages cooperation, but each side keeps its own AI and data rules.
- Start from systems: Inventory and classify every AI system before listing obligations.
- Shared controls save time: Risk assessments, model testing and oversight procedures often satisfy both sides.
- Gaps are your roadmap: Market-specific duties like EU conformity assessment get their own owners and deadlines.
- Keep it live: Review quarterly and watch the partnership’s working groups for changes.
Need Expert Help?
If this feels like a lot to manage alone, TecniForge can handle the heavy lifting. Our team specializes in custom software development and AI integration, including the logging, testing and documentation tooling that compliance evidence depends on. Get in touch with our experts.
Also read: EU-UK Digital Partnership: 7 Things Tech Firms Should Know After the Brussels Summit, our earlier coverage on why this matters today.
Quick test for your team: pick one AI feature and list every document you’d hand to an EU buyer and a UK buyer. How many appear on both lists? That number is your head start.
Discover more from TecniForge
Subscribe to get the latest posts sent to your email.