EU AI Act Compliance
Checklist for Businesses
A practical, phase-based readiness guide for founders and CEOs in finance and legal sectors. Map your systems, classify risk, and build a compliance roadmap that actually works.
The EU AI Act is not a suggestion. It is law.
The EU AI Act entered into force in August 2024 and is phasing in through 2026. Fines reach 35 million euros or 7% of global turnover. For finance and legal teams, the stakes are high: credit scoring, insurance pricing, legal document analysis, and recruitment tools all fall under high-risk or prohibited categories.
This checklist bridges the gap between regulatory theory and technical implementation. Each phase is ordered by urgency and designed for leaders who need to act, not just understand.
Note for non-EU companies: The Act applies extraterritorially. If your AI system is used in the EU or affects EU residents, you are in scope.
Four phases to compliance.
Map your AI footprint
You cannot comply with what you have not identified. Most companies underestimate how many AI tools they are already using. This phase is about discovery, not judgment.
- Inventory every AI system your company uses, builds, or resells. Include third-party tools, APIs, and embedded features.
- For each system, record: purpose, data inputs, decision outputs, and whether humans are in the loop.
- Determine your role: are you a provider, deployer, importer, or distributor for each system?
- Flag any system that makes or materially influences decisions about people. These are your high-risk candidates.
- Check if any system falls under prohibited practices (social scoring, biometric identification in public spaces, emotion recognition in workplaces).
Classify risk levels
The AI Act sorts systems into four buckets: prohibited, high-risk, limited risk, and minimal risk. Your obligations scale with the bucket. Getting this wrong means over-investing or under-protecting.
- Review Annex III of the AI Act against your inventory. High-risk categories include credit scoring, insurance pricing, recruitment, and legal document analysis.
- Check Annex I: if your system is already regulated under EU product safety law and uses AI, it is high-risk.
- Document the classification rationale for each system. Regulators will ask for this.
- Identify any general-purpose AI model you develop or fine-tune. These have separate transparency and systemic risk obligations.
- Reclassify after any major update or new use case. Risk classification is not a one-time exercise.
Build governance and documentation
High-risk systems need a paper trail: risk management systems, data governance, technical documentation, record-keeping, and human oversight protocols. Limited-risk systems need transparency disclosures. Start with the heavy lifting.
- Assign a responsible person or team for AI compliance. Solo founders often delegate this to their CTO or external counsel.
- Draft a risk management plan for each high-risk system. Cover identification, evaluation, mitigation, and monitoring.
- Document your training, validation, and testing datasets. The Act requires you to show data quality, representativeness, and bias mitigation.
- Write technical documentation that covers system architecture, performance metrics, known limitations, and intended use cases.
- Design human oversight mechanisms. For high-risk systems, a human must be able to understand, intervene, and override.
- Create logging and monitoring protocols. You need to track operational events, especially anomalies and errors.
Test, validate, and register
High-risk AI systems must pass conformity assessments before they go live. This is not a box-ticking exercise: it is a structured test of whether your system does what you claim, safely and fairly.
- Run pre-deployment testing that covers accuracy, robustness, fairness, and cybersecurity. Document results.
- Conduct a bias audit on protected characteristics relevant to your use case. Finance and legal have specific fairness expectations.
- Prepare an EU declaration of conformity. This is your signed statement that the system meets the Act's requirements.
- Register the system in the EU database for high-risk AI systems. You cannot deploy until this is done.
- For biometric or safety-critical systems, involve a notified body for third-party assessment.
- Set up post-market monitoring. Plan periodic reviews, incident reporting, and retraining triggers.
Risk categories at a glance.
| Risk level | Examples in finance & legal | Key obligation |
|---|---|---|
| Prohibited | Social scoring, manipulative AI, biometric ID in public | Must stop using immediately. Fines up to 7% of turnover. |
| High-risk | Credit scoring, insurance pricing, recruitment, legal doc analysis | Conformity assessment, risk management, human oversight, registration. |
| Limited risk | Chatbots, AI-generated content (deepfakes) | Transparency: users must know they are interacting with AI. |
| Minimal risk | Spam filters, recommendation engines, simple automation | No specific obligations. Voluntary codes encouraged. |