What is a high-risk AI system?

A system used in one of the areas of Annex III of the AI Act, such as access to education and assessment of learning, recruitment and management of workers, access to essential public services and benefits, and law enforcement, or one that is a safety component of a product covered by the product legislation in Annex I, or is itself such a product, and for which that legislation requires third-party conformity assessment. The intended purpose decides, not the technology: the same language model is a high-risk system when it screens job applications and is not when it summarises meeting notes. An Annex III system is nevertheless not high-risk if it poses no significant risk of harm, for example when it only performs a narrow procedural or preparatory task; profiling of natural persons is always high-risk, and a provider relying on the exception must document its assessment.

What must the deployer do?

Use the system in accordance with its instructions for use, assign human oversight to competent and trained persons, ensure the input data is relevant to the extent it controls it, monitor operation and report risks and serious incidents to the provider and the authority, keep the logs for at least six months unless the law provides otherwise, inform workers and their representatives before use at the workplace, inform people about whom the system makes or helps make decisions, and as a public authority register the use in the EU database.

What is a fundamental rights impact assessment and whom does it concern?

An assessment under Article 27 of the AI Act that bodies governed by public law, private entities providing public services, and deployers of systems assessing creditworthiness or insurance risk carry out before the first use of a high-risk system under Annex III, except critical-infrastructure systems under point 2 of Annex III. It describes the processes in which the system is used, the period and frequency of use, the groups of people likely to be affected, the risks of harm, how human oversight is implemented and the measures to take if the risks materialise, and its results are notified to the market surveillance authority.

From when do the obligations apply?

Regulation (EU) 2026/1744 postponed the start of application. For high-risk systems under Annex III the obligations apply from 2 December 2027 and for those under Annex I from 2 August 2028. Systems put into service before the application date are covered only if their design changes significantly, but systems intended for use by public authorities must comply by 2 August 2030 at the latest.

High-risk AI: what a deployer must do before December 2027

Articles 26 and 27 of the AI Act: human oversight, logs, informing workers, a fundamental rights assessment for public bodies. They apply from 2 December 2027 and 2 August 2028.

In brief

The AI Act classifies some AI systems as high-risk on the basis of their intended purpose, and sets their deployers obligations of their own in Article 26: use in accordance with the instructions, human oversight, input data quality, monitoring and reporting, retention of logs, informing workers and the people affected by decisions, and registration for public authorities. Public bodies and providers of public services must additionally carry out the fundamental rights impact assessment of Article 27 before first use, critical-infrastructure systems excepted. Regulation (EU) 2026/1744 moved the start of application of these obligations to 2 December 2027 and 2 August 2028. The time is meant for preparing, not for waiting.

What makes a system high-risk

The risk class comes from the intended purpose. Article 6 designates two sets of systems as high-risk. The first is systems that are a safety component of a product covered by the product legislation listed in Annex I, or are themselves such a product, for example in machinery and medical devices, and for which that legislation requires third-party conformity assessment; both conditions must be met. The second is the areas of use in Annex III, of which the most relevant to a municipality, a wellbeing services county or an ordinary employer are access to education and assessment of learning, recruitment, management of workers and employment decisions, evaluating access to essential public services and benefits, and law enforcement and the administration of justice.

The Annex III list is not absolute. Under Article 6(3), a system referred to in the annex is not high-risk if it does not pose a significant risk to health, safety or fundamental rights and does not materially influence the outcome of decision-making: for example where it only performs a narrow procedural task, improves the result of a previously completed human activity, detects decision-making patterns or deviations to support human review, or performs a preparatory task. Profiling of natural persons is always high-risk, however, and a provider relying on the exception must document its assessment before placing the system on the market.

The same technology is therefore a different thing in different use. A language model that summarises meeting notes is not a high-risk system. The same model scoring job applications or suggesting whose benefit application is checked first is. Telling these apart is level 1 of our skills catalogue, required of everyone: knows that some systems require specific measures before use.

The deployer's obligations in Article 26

Article 26 is written for the deployer, the organisation that uses the system in its own operations. Its paragraphs are in effect a checklist.

Paragraph	Obligation	What it means in practice
26(1)	Technical and organisational measures to ensure the system is used in accordance with its instructions for use	The provider's instructions are read and compliance organised; use contrary to the instructions shifts responsibility to the deployer
26(2)	Human oversight is assigned to persons with the necessary competence, training and authority, and they are given support	Named people with the right to override the system's output, and their training
26(4)	Input data is relevant and sufficiently representative to the extent the deployer controls it	What is fed into the system and where it comes from
26(5)	Operation is monitored; a risk is reported to the provider and the market surveillance authority and use is suspended; a serious incident is reported immediately	An owner for monitoring and a reporting route
26(6)	Automatically generated logs are kept for a period appropriate to the purpose, at least six months, unless Union or national law, in particular data protection law, provides otherwise	Log retention, agreed with the provider
26(7)	An employer informs workers' representatives and the affected workers before putting the system into use that they will be subject to a high-risk system	Consultation before use, not after
26(8)	A public authority registers the use in the EU database and must not use an unregistered system	A check in procurement
26(9)	The provider's information under Article 13 is used for the data protection impact assessment	AI as part of the impact assessment, not separate from it
26(11)	People are told that they are subject to a high-risk system when it makes or helps make decisions about them	A notice to the customer, the applicant, the pupil
26(12)	Cooperation with the competent authorities

The list shows that most of the obligations are organisation, not technology: who oversees, who monitors, who is told, where the logs are. At level 3 of our skills catalogue the head of an area implements these in that area, human oversight, logging and informing users, and at level 4 the AI officer is responsible for compliance as a whole.

The fundamental rights impact assessment in the public sector

Article 27 adds one obligation for the public sector. Deployers that are bodies governed by public law or private entities providing public services, and deployers of systems that assess creditworthiness or life and health insurance risk, must assess the possible impact on fundamental rights of using a high-risk system under Annex III before putting it into use, except for the critical-infrastructure systems in point 2 of Annex III, such as safety components in water or energy supply.

The assessment contains a description of the processes in which the system is used, the period and frequency of use, the categories of people and groups likely to be affected, the specific risks of harm to them in light of the information the provider has supplied, the implementation of human oversight, and the measures to be taken if the risks materialise, including internal governance and complaint mechanisms. The obligation applies to the first use; in similar cases the deployer may rely on an earlier assessment, and the assessment is updated when any element changes. The results are notified to the market surveillance authority on the template drawn up by the AI Office.

Paragraph 4 ties the assessment to data protection: where an obligation has already been met through a data protection impact assessment, the fundamental rights assessment complements it. In practice a municipality's AI procurement therefore goes through one assessment process in which the data protection officer and the AI officer each do their part.

What the Omnibus postponed

The original AI Act would have applied the high-risk obligations to Annex III systems from 2 August 2026 and to Annex I systems from 2 August 2027. Regulation (EU) 2026/1744 justifies the postponement by delays in standards, common specifications and national authorities, and sets the new dates: for systems under Annex III, Sections 1, 2 and 3 of Chapter III apply from 2 December 2027, and for systems under Annex I from 2 August 2028.

The transitional provision in Article 111 adds that the regulation applies to high-risk systems placed on the market or put into service before the application date only if their design changes significantly. Providers and deployers of systems intended for use by public authorities must nevertheless comply by 2 August 2030 at the latest. For the public sector an old system is therefore not a permanent exemption but a deadline.

What the postponement did not change: AI literacy under Article 4, the prohibited practices of Article 5 and the transparency obligations of Article 50 are in force. They are covered in Article 4 of the AI Act and Article 50 of the AI Act.

What to do now

For a deployer the transition period is time for three things. First an inventory: which systems in use or in procurement fall within the Annex III areas. A recruitment screening tool, learning analytics and prioritisation in benefit processing are the most common finds. Then responsibilities: who is the human overseer, who monitors, where the logs are kept, and how workers and customers are informed. Finally the assessment: in the public sector, a fundamental rights impact assessment alongside the data protection impact assessment.

None of this requires waiting for December 2027. It does require that the organisation has a person who knows how to do it, and that is level 4 of our skills catalogue: is responsible for compliance, carries out the fundamental rights impact assessment and documents it. The AI officer's role as a whole is described in Who in the organisation is responsible for AI use?.