EU AI Act High-Risk AI Systems: EU Commission Publishes Draft Guidance

2026-05-22 · Source: Debevoise Data Blog · Field: Legal & Regulatory — Compliance & Risk Management, Regulatory Affairs & Government Relations · Depth: Intermediate, medium

Summary

The European Commission has released draft guidance for classifying "high-risk" AI systems under the EU AI Act, including practical examples for various categories. This guidance, open for stakeholder feedback until June 23, 2026, clarifies how the Commission interprets the Act's high-risk classification rules. Its publication follows a provisional agreement to delay compliance obligations: stand-alone high-risk AI systems (Annex III) must comply by December 2, 2027, and embedded systems (Annex I) by August 2, 2028. The high-risk designation is crucial as it triggers extensive compliance requirements for providers, covering areas like risk management, data governance, and transparency, while deployers face a more limited set of duties. The guidance emphasizes that classification hinges on the AI system's intended purpose, not potential misuse, and distinguishes between administrative support and substantive influence in decision-making.

Key takeaway

For legal professionals advising on AI strategy, you should immediately integrate the EU Commission's draft guidance into your AI governance policies and risk-classification workflows. This clarifies the "intended purpose" principle and the broad interpretation of high-risk categories, which are then narrowed by the Article 6(3) filter. Ensure your organization's AI tools are clearly positioned for their intended use, especially in employment contexts, to avoid unintended "high-risk" classifications. Prioritize purpose-built AI tools for sensitive applications to better insulate your deployer clients from provider-level compliance burdens.

Key insights

EU AI Act "high-risk" classification hinges on intended purpose and substantive influence, clarified by new draft guidance.

Principles

• AI system classification is provider-led, based on intended purpose.
• High-risk categories are broadly interpreted, then narrowed via Article 6(3) filter.
• Distinguish administrative support from substantive influence in AI system impact.

Method

AI system classification under EU AI Act involves provider assessment of intended purpose, considering all materials, and applying Article 6(3) filter for narrowing scope.

In practice

• AI systems materially influencing recruitment or employment decisions are high-risk.
• Purpose-built AI tools for high-risk uses can reduce deployer compliance burdens.

Topics

• EU AI Act
• High-Risk AI Systems
• AI System Classification
• Regulatory Compliance
• AI Governance
• Employment AI

Best for: CTO, VP of Engineering/Data, Executive, Legal Professional, Director of AI/ML, Policy Maker

Related on AIssential

• Is my use case a high-risk AI system? Applying the Commission’s guidelines and next steps — The EU AI Act's high-risk classification is strict, with narrow derogations and significant obligations for providers and deployers.
• The EU AI Act, Decoded — The EU AI Act establishes a phased, risk-based regulatory framework for AI systems used within Europe.
• The EU is rapidly rewriting the AI Act. What’s changed? — The EU AI Act is being revised to balance innovation with regulation, particularly for general-purpose AI, by clarifying definitions and adjusting timelines.
• The Digital Iron Curtain: How the EU AI Act Is Strangling European Innovation — The EU AI Act's high compliance costs and restrictive classifications are stifling European AI innovation and competitiveness.
• AI Ethics Board 2026: Who Decides, What It Controls — Effective AI ethics boards require binding authority, not merely advisory capacity, to ensure genuine governance and compliance.
• Another Omnibus package – no.’VII’ – this time on all things AI — The EU's Omnibus VII package simplifies AI Act application, reducing burdens and clarifying overlaps for high-risk AI products.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Debevoise Data Blog.