High-risk AI classification: the two routes, the Article 6(3) filter and the new dates

An AI system is high-risk in two situations. Either it is a product, or the safety component of a product, covered by the EU legislation listed in Annex I of the AI Act and the product must undergo third-party conformity assessment (Article 6(1)); or it is intended for one of the use cases listed in Annex III — biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice and democratic processes (Article 6(2)).

The second route has a filter (Article 6(3)): the system escapes classification if it does not pose a significant risk to health, safety or fundamental rights. The filter never works, however, when the system profiles natural persons. And since Regulation (EU) 2026/1744, the Digital Omnibus on AI, entered into force on 27 July 2026, the obligations for Annex III systems apply from 2 December 2027 instead of 2 August 2026.

On 19 May 2026 the European Commission published draft guidelines on high-risk classification — the first document to explain Article 6 in detail, with practical examples. They are not binding, and after the public consultation the final version had not been published at the time of writing. Until it is, the draft remains the clearest indication of how supervisory authorities will reason.

The two routes to high-risk

Article 6(1) — AI in regulated products.Two cumulative conditions: the AI system is a safety component of a product, or is itself the product, covered by the harmonisation legislation in Annex I; and the product is subject to third-party conformity assessment. Typical examples are medical devices, vehicles, aircraft, lifts, toys, radio equipment and machinery. Not every AI in a regulated product becomes high-risk: without third-party assessment, the system stays outside the regime. A smart thermostat or a connected washing machine usually does not qualify. The draft guidelines read “safety component” autonomously and fairly broadly, so the analysis has to be made case by case.

Article 6(2) — Annex III systems. This is the route that concerns most software companies. Annex III is not a list of technologies but a list of use cases: what counts is the intended purpose, as described in the documentation, marketing and instructions. A mismatch between what the system does in practice and how its purpose is described does not protect the provider from classification.

The eight areas of Annex III

The draft guidelines organise the analysis by the eight areas of Annex III. For the obligations that follow, see high-risk AI systems.

• Biometrics: remote biometric identification, biometric categorisation by sensitive attributes and emotion recognition. One-to-one biometric verification — unlocking a phone, confirming that a person is who they claim to be — is expressly excluded, while some uses, such as categorising people by race or sexual orientation, are prohibited outright by Article 5.
• Critical infrastructure: safety components in the supply of electricity, water, gas and heating, road traffic and critical digital infrastructure.
• Education and vocational training: admission, evaluation of learning outcomes, assessing the appropriate level of education, monitoring behaviour during tests.
• Employment and worker management: recruitment and selection (filtering CVs, ranking candidates, targeted job advertising), decisions on promotion and termination, allocating tasks on the basis of behaviour or personal traits, monitoring and evaluating performance.
• Essential services: credit scoring and creditworthiness assessment, risk assessment and pricing in life and health insurance, eligibility for essential public benefits and services, triage of emergency calls.
• Law enforcement, migration and border control, justice and democratic processes: areas reserved mainly to public authorities.

The Article 6(3) filter: not everything in Annex III is high-risk

This is probably the most important clarification in the draft guidelines. A system that falls within an Annex III use case is not high-risk if it does not pose a significant risk to health, safety or fundamental rights, including by not materially influencing the outcome of decision-making. One of four alternative conditions is enough:

• a)
  A narrow procedural task

  For example, turning unstructured data into structured data, sorting documents into predefined categories or detecting duplicates.

• b)
  Improving the result of a completed human activity

  For example, flagging errors or inconsistencies in a decision already taken.

• c)
  Detecting decision-making patterns

  Detecting patterns or deviations from prior decisions, without replacing or influencing the human assessment absent proper human review.

• d)
  A preparatory task

  Performing a task preparatory to an assessment relevant to an Annex III use case.

The classic HR example: a system that only extracts data from CVs and makes it available to the recruiter may qualify for the exception. A system that scores candidates, ranks them or recommends who to invite to interview materially influences the decision — and remains high-risk.

Relying on the exception? Document it before launch

Under Article 6(4), a provider that considers its Annex III system not to be high-risk must document that assessment before placing it on the market or putting it into service, register the system in the EU database, and provide the documentation to national authorities on request. The Digital Omnibus kept the registration obligation, although the Commission had proposed removing it, and reduced the information required. Misclassification aimed at avoiding the requirements can be sanctioned. The exception is not a shortcut but a legal assessment, to be set out in writing with arguments that can be checked.

The dates after the Digital Omnibus

Regulation (EU) 2026/1744 was published in the Official Journal on 24 July 2026 and has been in force since 27 July 2026. The relevant dates:

• 2 August 2026 — the Article 50 transparency obligations apply: telling people they are interacting with AI, marking generated content, labelling deepfakes;
• 2 December 2026 — machine-readable marking for generative systems already on the market before 2 August 2026, and the new prohibited practices added by the Omnibus;
• 2 December 2027 — the full obligations for Annex III high-risk systems;
• 2 August 2028 — the obligations for AI in products regulated under Annex I.

The prohibited practices (Article 5) and the obligations for general-purpose AI models have applied since February and August 2025 respectively and are not affected by the postponement.

What it means for companies

The test is a two-step analysis. Does the system fall within an Annex III use case, or on the Annex I product route? If so, does it pose a significant risk to health, safety or fundamental rights, or does it qualify for the Article 6(3) filter? The postponement to December 2027 is not a reason to pause: the inventory of AI systems, their classification and the documentation of any exception do not depend on harmonised standards, and they are the starting point of every compliance programme. For a first orientation, a free AI risk classifier is available in Romanian.

Frequently asked questions