Conformity Assessment For High-Risk AI Systems

Drew & AI Comply HQ™ Team · July 2, 2026

A high-risk AI system must pass a check before it can be sold or used in the EU. The check shows it meets the EU AI Act's rules. It is called a conformity assessment (Article 43). The company that builds the system and offers it under its own name does the check. The Act calls that company the provider.

For 7 of the 8 high-risk areas the Act lists, the provider runs the check itself. No outside body is involved. The duty applies from December 2, 2027 for those listed uses, like hiring and credit scoring (Annex III). For AI built into products like medical devices (Annex I, Section A), it applies from August 2, 2028.

Which AI systems need a conformity assessment?

Only high-risk AI systems. These are systems where mistakes could seriously hurt people, and the Act lists them. There are two ways in (Article 6):

A listed use (Annex III). Examples are hiring and managing staff, credit scoring, and life and health insurance pricing. School admissions and grading, and access to public benefits, are on the list too. So are biometrics, critical infrastructure, law enforcement, migration, and justice.
A regulated product (Annex I). The AI is a product, or a safety part of a product, that already needs an outside safety check. Medical devices, toys, and lifts are examples.

An Annex III system can fall outside the high-risk group. That happens when it poses no significant risk of harm, for example because it does not meaningfully change the outcome of a decision. It must also fit one of four cases, like doing only a narrow or preparatory task (Article 6(3)). A system that profiles people is always high-risk. If you rely on this, write down why before launch and register the system in the EU database (Articles 6(4) and 49(2)). Our guide Is your AI system high-risk? walks through the test.

Which route applies to your system?

The Act sets the route, based on the kind of system. Some routes use an independent testing company approved by an EU country (called a notified body).

Your system	Route	Notified body?
Annex III, points 2 to 8: critical infrastructure, education, work, essential services, law enforcement, migration, and justice	Internal check (Annex VI)	No
Annex III, point 1 (biometrics), built to the official EU standards in full	Internal check (Annex VI) or notified body (Annex VII), your choice	Optional
Annex III, point 1 (biometrics), without those standards	Notified body (Annex VII)	Yes
Annex I, Section A product, like a medical device	The product law's own check, with the AI Act rules added	As the product law sets

If your Annex I, Section A product is also an Annex III use, follow the product law's route (Article 43(3)).

Some systems are supervised by the EU AI Office instead of national authorities. One example is a system built on the provider's own general-purpose AI model, with some exceptions. Another is AI inside a very large online platform or search engine (Article 75(1)). If such a system needs a notified body, you apply to the European Commission. A notified body then does the check on the Commission's behalf (Article 75(1e)).

What does the check cover?

Every route checks the same rules, set out in Articles 9 to 15:

Risk management (Article 9): finding and reducing the harm the system could cause
Data governance (Article 10): training and test data that fit the purpose
Technical documentation (Article 11 and Annex IV): a full record of how the system works and was tested
Record-keeping (Article 12): automatic logs of what the system does
Instructions for use (Article 13): clear information for the businesses that use it
Human oversight (Article 14): a person can watch the system, step in, and stop it
Accuracy, robustness, and cybersecurity (Article 15): the system performs well, copes with errors, and resists attacks

The internal check (Annex VI) has 3 parts. First, you confirm your written processes for building and running the AI (called a quality management system) meet Article 17. Next, you confirm your technical documentation shows the system meets the rules. Last, you confirm the system was built, and is monitored, the way that documentation says.

On the notified body route (Annex VII), the body reviews your quality management system and your technical documentation. If both pass, it issues a certificate. The certificate lasts up to 4 years for Annex III systems and up to 5 years for Annex I systems. You can renew it after a new review (Article 44). The body also checks your quality management system at regular intervals while the certificate is valid.

What do you do after you pass?

Sign an EU declaration of conformity (Article 47). It states that the system meets the rules. Keep it for 10 years.
Add the CE marking (Article 48). AI delivered online uses a digital CE marking. The marking must be quick to reach from the interface or through a machine-readable code. If a notified body was involved, its ID number goes next to the marking.
Register Annex III systems in the EU database (Article 49). Do it before the system is sold or used. Critical infrastructure systems are registered with the national authority instead.
Keep your records. Technical documentation and quality records stay available for 10 years (Article 18). Logs under your control stay for at least 6 months (Article 19).
Keep watching the system. Collect data on how it performs (Article 72) and report serious incidents (Article 73).

What if you change the system later?

If you change the system in a big way (called a substantial modification), you run a new conformity assessment (Article 43(4)). This applies even if the system stays with the same customer.

Some systems keep learning after launch. Changes you planned ahead and described in your technical documentation do not count as a big change.

When do you need it done?

The duty to finish the check before the system is sold or used applies from:

December 2, 2027 for Annex III systems
August 2, 2028 for Annex I, Section A systems. Machinery, vehicles, and other Section B products get their AI rules through their own EU laws (Article 2(2)).

Both dates come from the Digital Omnibus on AI (Regulation (EU) 2026/1744). It moved them in July 2026 (Article 113).

Systems already on the market before your date follow a different rule. They only have to comply if their design changes significantly after that date. AI used by public authorities must comply by August 2, 2030 either way (Article 111(2)).

Work back from your date. Start with the technical documentation, because every route checks it.

Is there help for smaller companies?

Yes, but there is no exemption. Small and medium-sized companies (SMEs) can fill in a simplified technical documentation form (Article 11(1)). The Digital Omnibus opened the form to companies just above that size. The Act calls them small mid-caps (SMCs). Notified bodies must accept the form.

Your quality management system can match the size of your company (Article 17(2)). SMEs also pay lower notified body fees (Article 62(2)).

What happens if you skip it?

Selling or using a high-risk system without the check breaks a provider duty (Article 16(f)). The fine for that can reach 15 million EUR or 3% of worldwide annual turnover (Article 99(4)). The higher of the two is the cap. For SMEs, and since July 2026 for SMCs, the cap is whichever is lower (Article 99(6) and (6a)).

Wrong or misleading answers to a notified body or an authority have their own cap. That cap is 7.5 million EUR or 1% of worldwide annual turnover, whichever is higher (Article 99(5)). Authorities can also require fixes, or take the system off the market (Article 79). Our guide to EU AI Act fines and penalties has the full picture.

Frequently asked questions

We use an AI system from a vendor. Is the conformity assessment our job?

Usually not. It belongs to the provider. As the business using the system (called a deployer), you have your own duties under Article 26. You become the provider if you put your name on the system or change it in a big way. The same goes if you change its purpose so it becomes high-risk (Article 25(1)). Our deployer obligations checklist lists your duties.

Can a notified body help us prepare?

No. Notified bodies must stay independent, so they cannot also act as your consultant (Article 31(5)). A consultant or compliance software can help you prepare. The notified body only checks the result.

Does our chatbot need a conformity assessment?

Not unless it is used for a high-risk purpose. A chatbot's main duty is to tell people they are talking to an AI (Article 50). That duty has applied since August 2, 2026.

Do general-purpose AI models need one?

No. Providers of general-purpose AI models, like the large language models behind chatbots, have their own duties (Articles 53 and 55). Those have applied since August 2, 2025. If you build a high-risk system on top of one, your system still needs its own conformity assessment. Our page on general-purpose AI model duties covers them.

Update: the Digital Omnibus is now law (October 7, 2026)

In July 2026 the EU changed some AI Act dates. The change is called the Digital Omnibus on AI (Regulation (EU) 2026/1744). It has been law since July 27, 2026. The dates in this article follow it.

What changed for conformity assessment:

High-risk rules start later (Article 113). AI used for hiring, credit scoring, education, and the other listed uses (Annex III) must comply from December 2, 2027. AI built into products like medical devices and toys (Annex I, Section A) follows on August 2, 2028. Machinery, vehicles, drones, and other Section B products get their AI rules through their own EU laws (Article 2(2)).
Small mid-caps get SME help (Articles 11 and 99). They can use the simplified technical documentation form. They also get the lower fine cap, except for the banned practices.
The EU AI Office supervises some systems itself (Article 75). That includes systems built on the provider's own general-purpose AI model, and AI inside very large online platforms. If one of them needs a notified body, the Commission organizes the check (Article 75(1e)). Annex I products, critical infrastructure, justice systems and some law enforcement, border and financial systems stay with national authorities (Article 75(1)(a)).
Fewer product features count as safety parts (Article 6(1a) and (1b)). AI used only for convenience, user help, or better performance is not a safety component. AI that could put health or safety at risk if it went wrong still is.

To see which dates apply to you, run the free risk check.

Sources

This page is informational content, not legal advice. Talk to a qualified lawyer about your specific situation.