EU AI Act update, 9 May 2026: current law remains the baseline. The Digital Omnibus provisional agreement would move many high-risk AI obligations to 2 Dec 2027 and product-integrated high-risk AI rules to 2 Aug 2028 if formally adopted. Track statusEU AI Act update: current law remains the baseline. Digital Omnibus dates apply only if formally adopted. Track status
Draft-guidance source update · 19 May 2026

EU AI Act Article 6(3) Filter Decision Record

An Article 6(3) filter decision record helps a team document the narrow filter rationale after the High-Risk Classification Router has already pointed the system toward Annex III and possible Article 6(3) review. It does not create an exemption, certify compliance, or replace counsel.

Source basis · last reviewed 20 May 2026

The European Commission published draft guidelines on high-risk AI classification on 19 May 2026. Use the High-Risk Classification Router first if you have not yet selected the Annex I, Annex III, or Article 6(3) review path. This page is only for building a working Article 6(3) filter record after an Annex III route is in scope. It is draft-guidance-based operational triage, not legal advice.

Decision rule

Use Article 6(3) as a record-building filter, not a shortcut.

The practical question is not “can we label this exempt?” The practical question is whether the system's task, profiling status, decision influence, and human assessment boundary can be documented clearly enough for review.

Task type

Record whether the AI system performs a narrow procedural task, preparatory task, post-human improvement, or pattern detection. Broad evaluation or scoring needs escalation.

Material influence

Record whether the output ranks, scores, excludes, recommends, prioritises, or otherwise shapes a meaningful decision about a person.

Profiling and evidence

Record whether profiling is involved and whether vendor instructions, intended purpose, logs, and oversight evidence support the claimed boundary.

Browser tool

Build the Article 6(3) filter decision record.

Use redacted facts. Do not enter personal data, trade secrets, employee records, patient data, privileged legal advice, production credentials, or client-confidential information.

1. System and role

Use an internal ID or redacted system name.

2. Annex III and task route

Select every area that reasonably applies.

3. Filter blockers and evidence

Output is a working record for internal review. It is not legal advice, conformity assessment, certification, or proof of compliance.

FAQ

Article 6(3) filter questions

These answers are designed for working records and vendor review, not final legal determinations.

No. The Article 6(3) Filter Decision Record gives an operational triage record, not a legal exemption decision. Use the output to document assumptions, evidence gaps, vendor questions, and review owners before qualified legal or regulatory review.

Use the Article 6(3) filter record after an AI system has been routed near an Annex III high-risk use case and you need to test whether the task may be narrow, procedural, preparatory, post-human improvement, or pattern detection without material decision influence.

Profiling is a serious escalation signal for Article 6(3) filter review. If the system profiles natural persons, do not rely on a filter rationale without legal and privacy review, vendor documentation, and a clear record of the system's intended purpose and actual workflow.

Keep the system name, intended purpose, Annex III area, profiling answer, material-influence answer, task-type rationale, human assessment note, vendor evidence status, reviewer name, review date, and a written explanation of why the filter rationale appears strong, weak, or incomplete.

The High-Risk Classification Router identifies whether an AI system should be reviewed through Annex I, Annex III, or the Article 6(3) filter path. This page is the next record-building step when the router result indicates possible Article 6(3) filter review.

This tool is designed to run the assessment logic in your browser. Do not enter confidential, personal, client-sensitive, privileged, trade-secret, employee, patient, credential, or production data. Use working names, internal IDs, or redacted summaries for internal review.

Source basis: Regulation (EU) 2024/1689 and European Commission draft high-risk classification guidelines published on 19 May 2026. Last reviewed: 20 May 2026. Use qualified legal, privacy, security, or sector counsel for final decisions.