Practical Compliance Guide

EU AI Act Compliance Checklist for SMBs (August 2026)

A practical, step-by-step checklist covering every EU AI Act obligation relevant to small and medium businesses — from prohibited practices to AI literacy. Work through each section in order.

10 min read·Last updated June 2026

Key Deadlines

2 Feb 2025Prohibited AI practices banned (Article 5) — PASSED
2 Aug 2025General-purpose AI model obligations apply — PASSED
2 Aug 2026High-risk AI system obligations (Annex III) + AI literacy enforcement
2 Dec 2030Legacy high-risk systems deployed before Aug 2026 must comply

Step 1: Determine If You Are Affected

The EU AI Act applies if your company is a provider or deployer of AI systems that are placed on the EU market or otherwise affect EU residents. This is true regardless of where your company is headquartered.

Your company is based in the EU or EEA

Your company is outside the EU but sells to EU customers

Your company is outside the EU but its AI outputs affect EU residents (even indirectly)

Your staff use AI tools (ChatGPT, Copilot, Gemini, AI-powered SaaS, etc.) in their work

If yes, you are a deployer and Article 4 AI literacy applies to you.

Your company develops, fine-tunes, or markets AI systems to others

If yes, you are a provider and additional obligations apply.

Outcome: If any of the above apply, continue with the checklist. If none apply, the EU AI Act does not currently affect you — but monitor for changes.

Step 2: Prohibited Practices Check (Article 5)

Before anything else: check whether any AI system you operate or plan to operate falls into a prohibited category. These are banned outright — no compliance programme can make them legal.

You do NOT use AI that performs real-time remote biometric identification in publicly accessible spaces

Narrow law-enforcement exceptions exist; seek legal advice if you think one applies.

You do NOT use AI that detects emotions in workplaces or educational institutions

This includes workplace productivity sentiment detection and student emotion monitoring.

You do NOT use AI that categorises individuals by race, ethnicity, political opinions, religious beliefs, or sexual orientation from biometric data

You do NOT use AI that exploits psychological vulnerabilities or subconscious manipulation to alter behaviour

You do NOT use AI to create social scoring systems (government or private entities assigning scores that determine access to services)

You do NOT use AI to predict individual criminal behaviour based solely on profiling

Outcome: If you ticked all of the above: you are clear of prohibited practices. If you could not tick any item, stop and take immediate legal advice — continued use risks fines up to €35 million or 7% of global annual turnover.

Step 3: Classify Your AI Systems

For each AI system you operate, determine its risk level. Use our free classifier tool or follow these questions.

You have created a list of all AI systems your business uses or provides

Include embedded AI features in SaaS tools (e.g. AI-powered recruitment screening in your HR platform).

For each system, you have identified whether it operates in an Annex III domain: employment, education, financial access, healthcare, law enforcement, border control, public services, or critical infrastructure

If yes to any, that system is high-risk.

For each system, you have identified whether it performs biometric processing (not just basic face unlock)

For each system that interacts with users, you have checked whether it clearly identifies as AI

If not, a transparency obligation applies.

For each system that generates content depicting real people, you have checked whether labelling is in place

Deepfakes and synthetic voices of real people require disclosure labelling.

Outcome: Document the risk classification of each system. Keep this record for at least 10 years for high-risk systems.

Step 4: High-Risk AI System Compliance (if applicable)

If you identified any high-risk AI systems in Step 3, these additional obligations apply. If all your systems are minimal or limited risk, skip to Step 5.

You have implemented a risk management process covering identification, evaluation, and mitigation of foreseeable risks

Must cover the entire AI lifecycle, from design to decommissioning.

Technical documentation is prepared and maintained (provider obligation)

Includes description of the system, training data, performance metrics, and known limitations.

Automatic logging of system operation is enabled with sufficient retention period

Deployers must maintain logs for at least 6 months; longer for some categories.

Human oversight mechanisms are implemented for all high-risk system outputs

Humans must be able to understand, monitor, and override the AI's output.

A conformity assessment has been conducted or obtained from the provider

The system has been registered in the EU AI database (deployers of high-risk AI must register)

The database is publicly accessible at the EU level.

A fundamental rights impact assessment has been conducted (deployers in public services, financial access, and employment)

Outcome: High-risk compliance requires meaningful investment in documentation and process. If you are using a third-party high-risk AI tool, ask your provider for their compliance documentation.

Step 5: Transparency Obligations (Article 50)

These obligations apply to any AI system that directly interacts with people or generates content — regardless of risk level.

Any AI chatbot or virtual assistant clearly discloses to users that they are interacting with AI

The disclosure must be clear and timely — ideally at the start of interaction.

AI-generated or AI-modified images, video, or audio depicting real, identifiable people are labelled as AI-generated

The C2PA standard is an acceptable technical labelling method.

Where users express a desire to interact with a human, this preference is not overridden without their consent

AI-generated content used for commercial communication or advertising is labelled

This includes AI-generated social media posts published by your business.

Outcome: Transparency obligations are low-cost to implement. Most can be addressed with a clear disclosure message in your product or content.

Step 6: Article 4 AI Literacy

This applies to ALL companies using AI tools — regardless of risk level. It has been in force since February 2, 2025.

You have created an inventory of AI tools used by your staff

You have written a brief AI usage policy for your organisation

Covers acceptable use, data handling rules, and who to contact with concerns.

Staff who operate or rely on AI outputs have completed AI literacy training

Minimum: what AI is, how your tools work, risks and limitations, privacy rules.

You keep a record of who completed training and when

An email confirmation, LMS record, or sign-in sheet all count.

New employees complete AI literacy training before working with AI tools

Training content is reviewed and updated at least annually

Outcome: AI literacy is the most universally applicable obligation. If you have done nothing else, start here.

Summary: Your Minimum Viable Compliance

For most SMBs that use AI tools but do not build high-risk systems, the most important actions are:

1

Check you are not using any prohibited AI practices (Step 2)

2

Create a list of AI tools your business uses

3

Write a short AI usage policy

4

Deliver Article 4 AI literacy training to staff who use AI, and keep records

5

If your AI systems interact with users, ensure they identify as AI

6

If you use any AI in employment, education, or financial decisions — seek legal advice about high-risk obligations

Official Resources

Classify Your AI System in 5 Minutes

Our free tool walks through your specific situation and tells you exactly which obligations apply.

Start Free Assessment