European Commission draft guidelines on classification of high-risk AI systems
Tag: S-2026-05-19-eu-ai-office-high-risk-draft Type: draft guidance + consultation Author(s): European Commission (Directorate-General CNECT / AI Office) Date of source: 2026-05-19 Date ingested: 2026-05-28 Authority weight: high — EU regulator’s own draft interpretive guidance under the AI Act. Raw file: S-2026-05-19-eu-ai-office-high-risk-draft.md. External URLs: https://digital-strategy.ec.europa.eu/en/library/draft-commission-guidelines-classification-high-risk-ai-systems ; https://digital-strategy.ec.europa.eu/en/news/commission-seeks-feedback-draft-guidelines-classification-high-risk-artificial-intelligence-systems
What it claims
On 19 May 2026 the European Commission issued draft guidelines clarifying the classification of high-risk AI systems under the AI Act, accompanied by a list of practical examples for stakeholder feedback. The Guidelines set out the Commission’s interpretation of concepts relevant to classification and provide worked examples of AI systems that should or should not be classified as high-risk. They are intended to support providers and deployers in self-assessing system classification. A targeted stakeholder consultation is open until 23 June 2026. The Guidelines are explicitly not legally binding but reflect the Commission’s interpretation and are stated to guide enforcement.
Notable quotes
“The Guidelines set out the Commission’s interpretation of certain concepts that are relevant for classification purposes and contain practical examples of AI systems that should or should not be classified as high-risk.” — Commission news (excerpt; full text not retrieved this run)
“Though not legally binding, they reflect the Commission’s interpretation and will guide enforcement.” — Commission news (excerpt)
What’s speculative vs. asserted
- Asserted: the 19 May 2026 issuance date; the 23 June 2026 consultation close; the inclusion of practical examples.
- ⚠️ Unconfirmed against the primary page (flagged 2026-06-05): the “not legally binding … will guide enforcement” characterisation does not appear on the Commission press release fetched directly; it came from a search-result summary. Verify against the draft guidelines PDF before relying on it.
- Speculative / forward-looking: the final wording of the Guidelines after consultation; how Member State competent authorities will weight the Guidelines in supervisory practice; how Commission enforcement will operationalise the worked examples.
Topics this feeds
Open questions raised
- Whether the draft examples align with banking-sector use cases (credit scoring, fraud detection, customer service AI).
- Whether the Final Guidelines will be published before the 2 August 2026 application date for high-risk obligations.
- The relationship between these Commission Guidelines and any future EBA / ESMA sector-specific elaboration.
Ingestion note
Search-derived from Commission news excerpts; direct fetch of the draft guidelines PDF was not completed in this scan. The publication fills the gap left by the missed statutory deadline of 2 February 2026 — see EU AI Office (European Commission) tracked changes. Confirm wording from the published draft before relying on this page for client deliverables.
2026-06-05 re-scan: the Commission press release was fetched directly (WebFetch). It confirms verbatim: the 19 May 2026 issuance; the 23 June 2026 feedback deadline; “The guidelines will support AI providers and deployers in assessing whether their system is high-risk or not”; “Under the AI Act, a limited list of use cases of AI systems are considered high-risk when they endanger health, safety or fundamental rights”; the draft is published on the AI Act Single Information Platform; and that these guidelines “will be complemented by other Commission guidelines to facilitate compliance with the obligations for high-risk AI systems”. The draft guidelines PDF itself remains unfetched.