Which obligations a high-risk system triggers
Anyone who provides or deploys a high-risk system must meet a catalogue of requirements that goes well beyond what applies to the lower risk classes. The most important obligations are:
Before a high-risk system is placed on the market, it undergoes a conformity assessment. The provider also registers Annex III systems in an EU database, and systems for critical infrastructure at national level. Providers must also maintain a quality management system and monitor the system’s performance after it has been placed on the market (post-market monitoring). Deployers, i.e. the companies that use the system, have their own obligations: they must follow the instructions for use, ensure human oversight and report relevant incidents.
Why this matters for SMEs
Many small and medium-sized companies assume that, as mere users of AI, they are not affected. That is a fallacy. As soon as a business uses a system that falls within an Annex III area, it becomes the deployer of a high-risk system, with its own obligations, even if the AI comes from an external provider.
The most common case in practice among SMEs is recruiting. Software that automatically filters CVs, evaluates candidates or creates a ranking is a high-risk system under Annex III. Anyone who buys such a solution must check whether the provider meets the conformity requirements and must anchor human oversight in their own process. Simply shifting responsibility to the manufacturer does not work.
The situation is similar for the creditworthiness assessment of natural persons, which plays a role in financing or instalment purchase contexts, for example. Here too, the high-risk classification applies as soon as an AI assesses creditworthiness and thus helps decide on access to a service.
A concrete example
An online retailer with 120 employees uses AI-supported applicant management that automatically sorts incoming applications by suitability and presents a pre-selection to the recruiters. This makes the retailer the deployer of a high-risk system within the meaning of Annex III (employment area). Once the obligations apply from 2 December 2027, this specifically means: the software provider must demonstrate a conformity assessment and have registered the system in the EU database. The retailer itself must ensure that the final personnel decision is made by a human, that the recruiters understand how the ranking is produced and that the processing is documented. If it uses the system without these precautions, it risks not only a fine but also discriminatory pre-selections that are legally challengeable. An exception applies to existing systems under the transitional rule in Article 111(2): if a unit of the same type and model was already placed on the market or put into service before 2 December 2027, the obligations only apply once the system’s design is significantly changed after that date (see “Current status and outlook”).
Common misconceptions
Several misconceptions persist around the term, and they can become expensive in practice.
“High-risk means prohibited.” Wrong. Prohibited practices are a separate, higher level (such as social scoring or emotion recognition in the workplace). High-risk systems are permitted but subject to requirements.
“This only affects the tech corporations that develop AI.” Wrong. The AI Act distinguishes between providers and deployers. Even those who merely use AI are deployers and have obligations.
“If the provider is compliant, I’m off the hook.” Only partly. The provider’s conformity is a prerequisite, but the deployer obligations (human oversight, correct use, reporting obligations) remain with the company using the system.
Providers and deployers: who bears which responsibility
The AI Act divides the obligations around a high-risk system between two roles, and understanding this separation determines who is liable in an emergency. The provider is whoever develops the system or has it developed and places it on the market under their own name. The heaviest obligations fall on the provider: conformity assessment, technical documentation, quality management, registration in the EU database and post-market monitoring of the system.
The deployer is whoever uses the system in their own operations. Its obligations are leaner, but by no means trivial: it must use the system in accordance with the instructions for use, ensure competent human oversight, keep input data relevant to the extent that it exercises control over it, retain the automatically generated logs and report serious incidents. There are also information obligations: employers inform workers’ representatives and the affected workers before putting a high-risk system into use at the workplace. Anyone who uses an Annex III system to make decisions about people must inform those affected, which in recruiting means the applicants. In certain constellations, a deployer can even become a provider itself, for example if it substantially modifies a purchased system or distributes it under its own name. The provider obligations then pass to it.
For SMEs, this division of roles is the key lever: anyone who only buys in high-risk AI should contractually ensure that the provider fulfils its obligations and, in parallel, anchor their own deployer obligations in the process. Only both together result in robust compliance.
How companies prepare
Even though the deadlines have been postponed, preparation is not a project that can be put off until the last moment. A structured approach has proven effective: first create a complete inventory of all AI systems in use, then assign each system to a risk class, then identify the few high-risk cases and go through the specific obligations for them. For every high-risk application, it pays to clarify early on whether the provider can demonstrate conformity and what human oversight specifically looks like in your own workflow. Anyone who lays these foundations will not face a mountain of work when the obligations start to apply, but a manageable remaining task.
Current status and outlook
The obligations for high-risk systems were originally to apply from 2 August 2026 (Annex III) and from 2 August 2027 (Annex I). The Digital Omnibus on AI, Regulation (EU) 2026/1744, postponed these dates: to 2 December 2027 for stand-alone Annex III systems and to 2 August 2028 for Annex I systems embedded in products. The Regulation was published in the Official Journal of the EU on 24 July 2026 and has been in force since 27 July 2026, so it is applicable law and no longer a draft. The background to the postponement is that the supporting technical standards and the supervisory infrastructure were not ready in time.
The core obligations from the table above remain in place; the postponement mainly provides more preparation time. In terms of content, the Omnibus clarified several points: it defines safety components more precisely, moved machinery to Section B of Annex I and allows small mid-cap enterprises, in addition to SMEs, to provide simplified technical documentation. A transitional rule applies to existing systems: high-risk systems placed on the market or put into service before the relevant date only fall under the obligations if their designs are subject to significant changes after that date. What matters is when the first unit of the same type and model came onto the market. Systems intended to be used by public authorities must comply by 2 August 2030 at the latest. For businesses with a high-risk application, the recommendation therefore remains the same: tackle the inventory and risk classification now instead of waiting for the postponed deadline. If you want to stay up to date, check the official sources regularly.
A detailed, continuously updated overview of the high-risk requirements can be found at the EU Artificial Intelligence Act project, which documents the text of the Regulation and its annexes.
Frequently asked questions
How do I know whether my AI system is high-risk?
Check whether the specific intended purpose lies in one of the Annex III areas (e.g. recruiting, credit checks, critical infrastructure) or whether the AI is a safety component in a product regulated under Annex I that requires a third-party conformity assessment. If either applies, the system is generally high-risk. By way of exception, an Annex III system may not be high-risk, for example if it only performs a narrow procedural task. The provider must document this assessment and must still register the system in the EU database. If the system performs profiling of natural persons, it is always high-risk.
Am I affected as a mere user?
Yes. Anyone who uses a high-risk system is a deployer and has their own obligations, above all human oversight, use in accordance with the intended purpose and the reporting of incidents.
When do the obligations apply?
Under the Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026), from 2 December 2027 for stand-alone Annex III systems and from 2 August 2028 for Annex I systems embedded in products. Systems that were already on the market before then are only covered if their design is significantly changed. For systems intended to be used by public authorities, 2 August 2030 applies at the latest.
What happens in the event of infringements?
The AI Act provides for fines of up to €15 million or 3 per cent of worldwide annual turnover for infringements of the obligations in the high-risk area. For SMEs and start-ups, the lower amount applies in each case, and since the Digital Omnibus on AI also for small mid-cap enterprises.
What is the first step towards compliance?
An inventory of all AI systems in use and their classification into the four risk classes. Only once it is clear which systems are high-risk can the specific obligations be assigned.
