High-Risk AI System

A high-risk AI system is a classification under the EU Artificial Intelligence Act, Regulation (EU) 2024/1689. It is distinct from an organization’s assessment that a use has serious consequences. Moving money does not by itself establish the classification; ranking job applicants falls within a specifically listed use. Absence of high-risk status does not exclude common duties, applicable transparency duties, or other laws. This entry describes the September 2026 text as amended by Regulation (EU) 2026/1744, including the distinction between classification, scope, and application dates.

Two routes in

Article 6 gives two ways for a system to be high-risk. The first is through product law: a system is high-risk when it “is intended to be used as a safety component of a product, or the AI system is itself a product” covered by the EU harmonization legislation listed in Annex I, and that product must undergo third-party conformity assessment. Machinery, medical devices, vehicles, and lifts are the kind of product meant. The 2026 amendment narrowed the safety-component idea: systems “solely used for non-safety related aspects of user assistance, performance optimisation, service efficiency, automation or convenience or quality control” are not safety components, while systems “the failure or malfunctioning of which would endanger health and safety” are.

Article 6(1c) excludes the third-party-assessment condition where that assessment concerns only specified risks unrelated to health and safety. Article 2(2) also gives Annex I Section B systems a separate scope: the full Chapter III obligation list below must not be applied to every listed product without checking its sectoral route.

The second route is a list. Article 6(2) states that “AI systems referred to in Annex III shall be considered to be high-risk”, and Annex III names eight areas: biometrics; critical infrastructure; education and vocational training; employment, workers’ management and access to self-employment; access to essential private and public services and benefits, which covers public assistance, creditworthiness of natural persons (with a financial-fraud exception), risk assessment and pricing for natural persons in life and health insurance, and emergency dispatch; law enforcement; migration, asylum and border control; and the administration of justice and democratic processes. Each area is itemized. Under employment, for instance, the Annex lists systems intended for “the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates”, and separately systems used for decisions on promotion, termination, task allocation, and the monitoring of workers.

The exception, and its exception

Being in an Annex III area does not settle the question. Article 6(3) provides that such a system is not high-risk “where it does not pose a significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the outcome of decision making”, and it applies that derogation where any of four conditions holds: the system “is intended to perform a narrow procedural task”; it “is intended to improve the result of a previously completed human activity”; it is intended to detect decision-making patterns or deviations from them without replacing or influencing a completed human assessment, absent proper human review; or it “is intended to perform a preparatory task to an assessment” relevant to an Annex III use. Then comes the limit on the limit: a system in an Annex III area “shall always be considered to be high-risk where the AI system performs profiling of natural persons.”

Applied to recruiting, a fictional pair shows the line. A screener that reads applications and ranks candidates materially influences the outcome and evaluates people; it is high-risk. A tool that extracts name, dates, and qualifications from a CV into a structured form, with no scoring, is a preparatory task and plausibly not. The Act does not leave that judgment informal. Article 6(4) requires a provider that considers an Annex III system not to be high-risk to “document its assessment before that system is placed on the market or put into service”, to register it, and to produce the documentation on request. Article 6(5) directs the Commission to publish guidelines with practical examples of systems that are and are not high-risk, and Article 6(6) allows the conditions to be amended by delegated act, which is why any reading of this paragraph carries a date.

What follows from the label

An organization developing or commissioning a system and putting it into service under its own name can be its provider even without selling it. Using it under its authority can also make it a deployer. Article 25 addresses rebranding an existing high-risk system, substantially modifying one so it remains high-risk, and changing a system’s intended purpose so it becomes high-risk. Determine the actual roles and changes; not every alteration to any AI system transfers high-risk provider duties.

Within the applicable high-risk route and dates, provider duties include risk and quality management, data governance, technical documentation, logging, instructions, oversight design, appropriate accuracy and cybersecurity, conformity assessment, and post-market monitoring. Registration and serious-incident reporting have the scopes specified in Articles 49 and 73. Deployers follow instructions, provide competent and supported oversight, monitor use, and check relevant input data under their control. Article 26 requires the applicable risk and serious-incident notifications, including to authorities rather than only the provider, and suspension where required. Logs under the deployer’s control are kept for an appropriate period of at least six months unless applicable Union or national law provides otherwise. Worker and affected-person notices have their own scope. Article 27 fundamental-rights assessments concern specified deployers of Article 6(2) systems, with Annex III point 2 excluded; they cover public-law bodies, private public-service providers, and the deployers of uses in points 5(b) and 5(c).

Article 113 sets the relevant Chapter III application dates at 2 December 2027 for Annex III and 2 August 2028 for Annex I. Article 111 must be read alongside them. For systems already marketed or put into service before the relevant date, paragraph 2 generally conditions application on significant design changes from that date, with an additional 2 August 2030 deadline for systems intended for public authorities and special provisions for certain large-scale systems. Record the deployment and change history rather than assuming all existing systems become fully covered automatically or remain exempt forever. Other obligations and laws can apply independently. South Korea’s high-impact category has its own criteria. The Invoice Gave the Instructions explains the wider regulatory layers; human oversight and automation bias discuss the conditions for effective supervision.

References:

• Regulation (EU) 2024/1689 (Artificial Intelligence Act), consolidated text of 27 July 2026, EUR-Lex, Articles 2, 3, 4, 6, 9 to 17, 25, 26, 27, 43, 49, 72, 73, 111 and 113, and Annex III.
• Regulation (EU) 2026/1744 (Digital Omnibus on AI), EUR-Lex, 8 July 2026.
• Center for Security and Emerging Technology, Framework Act on the Development of Artificial Intelligence and Establishment of Trust (translation), South Korea.