Article 6(3) provides that, by derogation from Article 6(2), an AI system referred to in Annex III shall not be considered 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 where one of four conditions is fulfilled: the system is intended to perform a narrow procedural task; to improve the result of a previously completed human activity; to detect decision-making patterns or deviations from prior patterns without replacing or influencing the previously completed human assessment without proper human review; or to perform a preparatory task to an assessment relevant for an Annex III use case. An Annex III system that performs profiling of natural persons shall always be considered high-risk, regardless of the four conditions. A provider relying on the filter must document the assessment before placing on the market or putting into service, and is subject to registration under Article 49(2).
What is the filter, and what is it not?
The filter is a derogation from Article 6(2) only. It does not apply to Article 6(1) product systems, it does not remove Article 5 prohibitions, and it does not remove Article 50 transparency duties or Chapter V model duties. It allows an Annex III system to be treated as not high-risk only where both the no significant risk test and one of the four conditions are met. The filter is therefore a narrow, evidence-based exception, not a general low-risk label. A provider that treats every assistive or preparatory function as filtered, without testing the risk and condition elements, misunderstands the provision. The filter must be argued, evidenced and documented, not asserted from a product description.
What does no significant risk of harm mean?
The system must 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. Material influence means the system's output shapes the decision in a way that matters to the outcome, even if a human formally makes the final decision. A system that prepares a shortlist, ranks candidates, or highlights cases for attention can materially influence the outcome, even where a human reviews the list. A system that performs a purely administrative step with no influence on the substantive decision is a different case. The record should describe the decision, the system's influence on it, the harm that could follow from error, and why that harm is not significant in the specific context, with evidence rather than assertion.
For the broader regulatory framework, see high-risk AI systems.
What are the four conditions?
The four conditions are alternatives: any one can be relied on, provided the no significant risk test is also met. The first is a narrow procedural task, such as converting a document format or checking completeness, without substantive evaluation. The second is improving the result of a previously completed human activity, such as correcting language in a decision already made. The third is detecting decision-making patterns or deviations from prior patterns, without replacing or influencing the previously completed human assessment without proper human review. The fourth is performing a preparatory task to an assessment relevant for an Annex III use case, such as gathering documents for a human assessment. Each condition must be matched to the actual function, and a system that performs several functions must be assessed function by function.
Why does profiling always defeat the filter?
Article 6(3) states that, notwithstanding the first subparagraph, an AI system referred to in Annex III shall always be considered high-risk where it performs profiling of natural persons. Profiling means automated processing of personal data to evaluate personal aspects, including work performance, economic situation, health, preferences, interests, reliability, behaviour, location or movements. Many Annex III systems perform profiling by design: recruitment screening, credit scoring, risk assessment and performance evaluation all evaluate personal aspects. A provider that relies on the filter must therefore address profiling directly and honestly. If the system profiles, the filter is unavailable, and the system remains high-risk. A record that does not mention profiling, where profiling is plainly present, is not a filter assessment.
What documentation and registration follow reliance on the filter?
Article 6(4) requires a provider who considers that an Annex III system is not high-risk under the filter to document its assessment before the system is placed on the market or put into service, and to provide that documentation to national competent authorities on request. The provider is also subject to the registration obligation in Article 49(2), which requires registration of the system in the EU database in the circumstances that article sets. The documentation should record the Annex III point, the no significant risk reasoning, the condition relied on, the profiling analysis, the evidence examined, and the reviewer. A filter conclusion without documentation and registration is incomplete, and a later authority request should be answerable from the file, not reconstructed from memory.
Continue with Article 6 classification routes.
What should not be claimed from the filter?
This guide explains the filter and its limits. It does not decide whether any particular system meets the no significant risk test, meets a condition, or avoids profiling. Those questions need the function, the decision affected, the data, and the human review design, read against the consolidated text and the Commission guidelines on Article 6 implementation, which Article 6(5) required by 2 February 2026 and which should be checked in their current form. Where the filter is the difference between a high-risk regime and no high-risk regime, confirmation should be sought from a competent authority or legal adviser before the system is placed on the market or put into service, because the cost of a wrong filter conclusion is borne by the affected people and by the provider.
Frequently asked questions
Does performing a narrow task automatically avoid high-risk status?
No. The narrow procedural task is one of four conditions, and the no significant risk test must also be met, including no material influence on the outcome, and profiling must be absent. All elements must be evidenced and documented.
Must a filtered system still be registered?
Yes, in the circumstances Article 49(2) sets. A provider relying on Article 6(3) is subject to the registration obligation in Article 49(2), and must document the assessment before placing on the market or putting into service and provide it to authorities on request.
Can the filter apply to a system that profiles people?
No. Article 6(3) provides that an Annex III system that performs profiling of natural persons shall always be considered high-risk, notwithstanding the filter conditions.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689 ↗AI Act Service Desk timeline ↗European Commission regulatory framework for AI ↗