In this article18 sections
- Who this checklist is for
- The EU AI Act compliance checklist at a glance
- 1. Build your AI system inventory
- 2. Decide your role: provider, deployer or both
- 3. Classify each system by risk
- 4. Rule out prohibited practices (Article 5)
- 5. Take AI literacy measures (Article 4)
- 6. Meet the transparency duties (Article 50)
- 7. Prepare the high-risk deployer duties (Article 26)
- 8. Do a FRIA if you are in scope, and a DPIA alongside
- 9. Collect vendor and GPAI documentation
- 10. Set up serious incident reporting
- 11. Track the timeline, including the Digital Omnibus
- 12. Keep an evidence file
- What getting it wrong can cost
- Run it every year: the yearly review
- Frequently asked questions
- References
An EU AI Act compliance checklist for an SME has twelve checks: list your AI systems, fix your role, classify each by risk, rule out banned uses, train staff (Article 4), meet transparency duties (Article 50), prepare high-risk duties (Articles 26 and 27), line up vendors, incident reporting and your DPIA, track the dates, and keep an evidence file.
TL;DR
-
Most SMEs are deployers: they use AI someone else built, with lighter duties than a provider.
-
Bans (Article 5) and AI literacy (Article 4) apply since 2 February 2025, transparency (Article 50) since 2 August 2026.
-
High-risk deployer duties start on 2 December 2027 for Annex III systems and 2 August 2028 for AI in Annex I products, after the Digital Omnibus moved them.
-
The table below is the whole checklist. Run it once now, then every year.
Who this checklist is for
You run an EU company that uses AI at work: a website chatbot, an AI feature in your HR software, Copilot on staff laptops. You do not train your own models.
"SME" means the EU definition: fewer than 250 staff and a turnover up to €50 million or a balance sheet up to €43 million, per the European Commission's SME definition. Size matters most for fines: for SMEs and start-ups, each ceiling in Article 99 is the lower of its two figures.
If you also sell AI under your own name, you are a provider too, with heavier duties than this list covers. This page is not legal advice.
The EU AI Act compliance checklist at a glance
| # | Check | Where it comes from | Applies from |
|---|---|---|---|
| 1 | Keep an inventory of every AI system you use | Basis for every check below | Now |
| 2 | Decide your role per system: provider, deployer or both | Article 3, Article 25 | Now |
| 3 | Classify each system by risk, including Annex III | Article 6, Annex III | Now (duties follow the class) |
| 4 | Confirm no system falls under a ban | Article 5 | 2 February 2025; two new bans 2 December 2026 |
| 5 | Take AI literacy measures for staff | Article 4 | 2 February 2025 |
| 6 | Meet transparency duties (deepfakes, emotion recognition, AI text, chatbots) | Article 50 | 2 August 2026 |
| 7 | Prepare high-risk deployer duties: instructions, oversight, logs, informing people | Article 26 | 2 December 2027 (Annex III), 2 August 2028 (Annex I) |
| 8 | Do a FRIA if you are in scope, and a DPIA under the GDPR | Article 27, GDPR Article 35 | FRIA with Article 26; DPIA already |
| 9 | Collect vendor and GPAI documentation | Articles 13, 26, 53 | Now, for the 2027 deadline |
| 10 | Set up serious incident reporting | Article 26(5), Article 73 | With Article 26 |
| 11 | Track the timeline, including the Digital Omnibus | Regulation (EU) 2026/1744 | Ongoing |
| 12 | Keep an evidence file and review it yearly | Supports every check | Now |
1. Build your AI system inventory
You cannot classify or disclose a system you do not know about, and AI features arrive quietly inside tools you bought years ago.
Check:
-
Every tool with an AI feature, per department, including free tools staff use on their own.
-
For each: name, vendor, purpose, users, data that goes in, an internal owner, and the date of the last review.
A spreadsheet is enough. Our guide to building an AI system inventory in 30 minutes walks through it.
2. Decide your role: provider, deployer or both
If you bought a tool and use it under the vendor's name, you are a deployer. If you developed it, or market it under your own name, you are a provider.
The line can move. Under Article 25, a deployer becomes the provider of a high-risk system if it puts its own name or trademark on it, makes a substantial modification to it, or changes a system's intended purpose so that it becomes high-risk.
Check: mark each inventory line as provider, deployer or both, and flag any system you have rebranded, retrained or repurposed.
Software companies often sit on both sides: see provider vs deployer obligations for SaaS.
3. Classify each system by risk
The AI Act has four risk levels, and the Commission says the vast majority of AI systems used in the EU are minimal or no risk. The ones to watch are in Annex III, which lists eight high-risk areas: biometrics, critical infrastructure, education, employment, essential services (such as credit scoring and life and health insurance pricing), law enforcement, migration, and justice and democratic processes.
For an SME, employment is the usual one: a tool that filters CVs, ranks candidates or evaluates staff is high-risk. Article 6(3) exempts some Annex III systems that pose no significant risk, but apply that narrow test with care.
Check:
-
Give each system a class (prohibited, high-risk, transparency duty, minimal) and write down why.
-
Recheck when a tool's use changes.
Read the 8 Annex III categories explained, or use the 30-minute risk classification self-assessment.
4. Rule out prohibited practices (Article 5)
Article 5 has banned a short list of uses since 2 February 2025. The ones an SME is most likely to meet: emotion recognition in the workplace or in education (except for medical or safety reasons), manipulative techniques that cause significant harm, and exploiting vulnerabilities due to age, disability or social or economic situation. Social scoring and biometric categorisation that infers sensitive traits are banned too.
From 2 December 2026, two more bans apply: non-consensual sexual deepfakes of an identifiable person, and child sexual abuse material.
Check: no tool infers the emotions of staff or candidates or scores people on social behaviour, and generative image tools have safeguards for the new bans.
Every ban is explained in our Article 5 guide.
5. Take AI literacy measures (Article 4)
Article 4, as amended by Regulation (EU) 2026/1744, requires providers and deployers to "take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf". It does not require any specific level of AI literacy per person.
The law sets no training hours or record-keeping duty, but the Commission's AI Literacy Q&A suggests keeping an internal record of trainings and other initiatives.
Check:
-
Everyone who uses AI at work has guidance that fits their tools and role; new hires get it before they start.
-
Staff in high-risk uses, such as recruitment, get more depth.
-
You log who did what, and when.
See what Article 4 means for European SMEs in practice.
6. Meet the transparency duties (Article 50)
Article 50 has applied since 2 August 2026. As a deployer, you must:
-
Disclose that image, audio or video content is artificially generated or manipulated when it is a deepfake (lighter disclosure for evidently artistic, satirical or fictional work).
-
Disclose AI-generated text you publish to inform the public on matters of public interest, unless a person reviewed it and someone holds editorial responsibility.
-
Inform people exposed to an emotion recognition or biometric categorisation system that it is in operation.
Telling people they are talking to an AI is the provider's duty (Article 50(1)). If you built your own chatbot, that is you.
Our Article 50 guide to chatbots and deepfakes has examples.
7. Prepare the high-risk deployer duties (Article 26)
If check 3 found a high-risk system, Article 26 is your main list, from 2 December 2027 for Annex III systems and 2 August 2028 for Annex I systems. As a deployer you must:
-
Use the system in line with the provider's instructions for use (26(1)).
-
Assign human oversight to people with "the necessary competence, training and authority" (26(2)).
-
Make sure input data you control is relevant and sufficiently representative (26(4)).
-
Monitor the system, and suspend use and notify the provider and authority if you believe it presents a risk (26(5)).
-
Keep the logs it generates under your control for at least six months, unless other law says otherwise (26(6)).
-
Inform workers' representatives and affected workers before using it at the workplace (26(7)).
-
Inform people that they are subject to the system when it makes or assists decisions about them (26(11)).
Check: each high-risk system has a named oversight owner who can override it, you know where its logs live, and staff have been informed or a date is set.
Recruitment is where most SMEs meet these duties: see what Article 26 means for HR.
8. Do a FRIA if you are in scope, and a DPIA alongside
A Fundamental Rights Impact Assessment under Article 27 is not for every deployer. It applies to bodies governed by public law, private entities providing public services, and deployers of AI for credit scoring or for life and health insurance pricing. A private employer using recruitment AI is not in that group. If you are in scope, you notify the market surveillance authority of the results.
The GDPR Data Protection Impact Assessment applies today when AI processing is likely to result in a high risk to people. Article 26(9) tells deployers to use the provider's information for it, and Article 27(4) lets a FRIA reuse parts of the DPIA.
Check: write down whether you fall in the Article 27 group and why, and keep a current DPIA for each high-risk processing of personal data.
Read DPIA vs FRIA: do you need both, and when.
9. Collect vendor and GPAI documentation
For a high-risk system, you need the provider's instructions for use to meet Article 26. For general-purpose AI models, such as those behind ChatGPT or Gemini, provider duties have applied since 2 August 2025. Article 53 makes those providers share documentation with providers of AI systems built on the model. If you only use a finished tool, you get what the vendor publishes or puts in its business terms.
Check:
-
High-risk tools: instructions for use, logging, and a clear answer on conformity by 2 December 2027.
-
Every AI vendor: a contract that allocates incident reporting, documentation and data processing.
-
GPAI tools: usage policies, model or system cards and the training-content summary, filed.
See what you can ask OpenAI, Anthropic and Google and the vendor due diligence checklist for AI chatbots.
10. Set up serious incident reporting
Under Article 73, the provider of a high-risk system reports a serious incident to the market surveillance authority within 15 days of becoming aware of it, 10 days in the event of a death, and 2 days for a widespread infringement.
Your duty as a deployer sits in Article 26(5): when you identify a serious incident, immediately inform first the provider, then the importer or distributor and the market surveillance authority.
Check: staff know how to flag an AI problem and to whom, you log incidents and near misses, and you have each high-risk provider's incident contact.
Read who reports what under Article 73.
11. Track the timeline, including the Digital Omnibus
The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026) changed the dates. As it stands:
-
2 February 2025: prohibited practices (Article 5) and AI literacy (Article 4).
-
2 August 2025: obligations for general-purpose AI model providers, and the penalties chapter.
-
2 August 2026: general application, including Article 50. National market surveillance authorities supervise, including Article 4.
-
2 December 2026: two new prohibitions, and the end of the grace period for machine-readable marking (Article 50(2)) for generative systems on the market before 2 August 2026.
-
2 December 2027: high-risk rules for Annex III systems, such as recruitment, HR and credit scoring.
-
2 August 2028: high-risk rules for AI in regulated products (Annex I).
In the Netherlands, the implementing law is still a draft, so no Dutch authority has AI Act powers yet. Read what changed on 2 August 2026 and what the Digital Omnibus changed for SMEs.
12. Keep an evidence file
If an authority asks, you must show what you did. Check that you can produce:
-
Your AI inventory with risk classes and the reasoning.
-
AI literacy records.
-
Your AI use policy, dated, with version history.
-
Incident logs.
-
Vendor documentation: instructions for use, conformity information, contracts.
-
Your DPIA, and a FRIA where Article 27 applies.
See what EU regulators actually want in an evidence file.
What getting it wrong can cost
Article 99 caps fines at €35 million or 7% of global annual turnover for prohibited practices, €15 million or 3% for operator duties such as Articles 26 and 50, and €7.5 million or 1% for misleading information to authorities. Article 4 has no EU ceiling; national law sets those penalties. Read how the fine amount is set, or estimate your ceiling with the AI Act fine calculator.
Run it every year: the yearly review
The AI Act does not prescribe a yearly review, but its duties are ongoing, and tools, staff and the law all change. Six areas carry most of the work:
-
Inventory. Has any tool changed function or purpose? A tool now used for recruitment scoring has moved into Annex III.
-
AI literacy. Were new hires and people in new roles trained? Does the content still fit your tools and the amended Article 4?
-
Policies. Does your AI use policy list every tool, name an owner per high-risk system and set an escalation path? Date each version.
-
Incident log. Look for patterns. An empty log usually means people are not reporting.
-
Vendors. Did they update documentation or report changes? Are contracts, DPIAs and any FRIA still accurate?
-
Audit readiness. Run a mock pull of the evidence file. If it takes longer than two hours, fix the filing now.
Your 12-month calendar
| Month | Task |
|---|---|
| January | Refresh the AI inventory and risk classes |
| February | Review AI literacy records and training content |
| March | Update the AI use policy and get sign-off |
| April | Review the incident log and close open actions |
| May | Reassess vendors, contracts, DPIAs and any FRIA |
| June | Mock document pull of the evidence file |
| July | Check for changes in EU law and your national implementing law |
| August | Check transparency notices on customer-facing tools |
| September | Train new hires and staff who changed roles |
| October | Spot check the inventory for new tools |
| November | Prepare the high-risk duties that start on 2 December 2027 |
| December | Set next year's owners and budget |
Frequently asked questions
Is there an EU AI Act compliance checker?
Yes. Our free compliance check is 15 questions and takes about two minutes, and you see your score straight away. It is a starting point, not a legal opinion.
Where can I find a checklist for AI Act compliance readiness?
The table at the top of this page: twelve checks, each with its article and start date.
What should an AI compliance checklist for SMEs include?
The twelve checks above. If you have no high-risk system, checks 7, 8 and 10 shrink to a yearly recheck, and the rest still apply.
Does the EU AI Act apply if we only use AI tools?
Yes. Using AI at work makes you a deployer. Articles 4 and 5 already apply, Article 50 applies if you use deepfakes or emotion recognition or publish AI text on public-interest matters, and high-risk duties apply only to Annex III systems.
Can an AI compliance checklist be automated?
Partly. Inventory tracking, training logs, log retention and reminders can run in software. Classifying a system, judging whether a use is banned and assigning oversight still need a person who knows the tool and its context.
When do the high-risk rules apply?
From 2 December 2027 for Annex III systems and 2 August 2028 for AI in Annex I products, the dates set by the Digital Omnibus on AI.
References
Not sure where your company stands?
Answer 15 questions about how your team uses AI. You get a compliance score and the gaps to close first, as a PDF.