EU delivers high-risk AI guidelines

The European Commission on May 19 published draft guidelines to help companies, public bodies and regulators decide when an AI system falls into the “high-risk” category under the EU AI Act. The package is aimed at Article 6 of the law, which sets the classification rules, and includes practical examples for different use cases. A separate Commission consultation, opened on May 8, seeks feedback on draft guidance for the AI Act’s transparency obligations. ### Which part of the AI Act is the Commission trying to clarify? Article 6 of the AI Act is the provision that determines whether an AI system is high-risk. The Commission said its draft guidelines are intended to support providers, deployers and market-surveillance authorities in assessing that status and to promote “uniform application and effective enforcement” of Article 6. (digital-strategy.ec.europa.eu) The Commission split the material into general principles and separate guidance for Annex I and Annex III cases. Annex I covers AI tied to products regulated under existing EU product-safety laws and subject to third-party conformity assessment. Annex III covers standalone use cases in areas such as employment, education, law enforcement and migration that the AI Act treats as high-risk, subject to conditions and exclusions in the law. (digital-strategy.ec.europa.eu) ### How does the draft tell companies to decide if a system is high-risk? The Commission’s draft sets out a three-step assessment structure. Companies are first asked to identify whether the software is an AI system under the Act, then determine whether it falls within Annex I or Annex III, and then assess whether any exemptions or qualifications apply before treating it as high-risk. That structure is reflected in the Commission’s description of the guidelines and the way the documents are organized around Article 6 implementation. (digital-strategy.ec.europa.eu) The examples matter because the classification question triggers the AI Act’s heaviest compliance duties. Once a system is high-risk, providers and deployers face requirements on risk management, data governance, record-keeping, transparency, human oversight, accuracy, robustness and cybersecurity under the broader framework of the law. The Commission has also linked the guidelines to its standardisation work under the Act. (digital-strategy.ec.europa.eu) ### What is in the separate transparency guidance? The Commission on May 8 published draft guidelines on Article 50, which sets transparency obligations for certain AI systems. The draft says it is meant to give practical guidance to competent authorities, providers and deployers so they can comply with those obligations “in a consistent, effective and uniform manner.” (digital-strategy.ec.europa.eu) Article 50 requires, among other things, that providers inform people when they are interacting with an AI system in certain contexts and add machine-readable markings to help detect AI-generated or manipulated content. The Commission’s consultation notice also says deployers must inform people when they are exposed to some AI-generated or manipulated material. (digital-strategy.ec.europa.eu) ### Who is the Commission asking to respond, and by when? The Commission said stakeholders including AI providers, developers, businesses, public authorities, academics, researchers and citizens can comment on the high-risk classification draft until June 23, 2026. The consultation page says the guidelines include clarifications on the relevant AI Act provisions and practical examples showing how classification should be assessed across use cases. (digital-strategy.ec.europa.eu) The transparency consultation opened on May 8 and closes on June 3, 2026. The Commission said that draft takes into account input from earlier consultations and is being published for feedback ahead of adoption. ### What comes next for companies selling AI into Europe? (digital-strategy.ec.europa.eu) The AI Act Service Desk says the Commission was required to issue Article 6 guidance after consulting the European Artificial Intelligence Board and no later than February 2, 2026. The May 19 publication means companies now have draft text and examples to work from while the Commission collects comments and prepares final guidance. (digital-strategy.ec.europa.eu) The next dated milestones are June 3 for feedback on the transparency draft and June 23 for comments on high-risk classification. Both consultations are hosted by the Commission’s digital policy arm, which has published the draft texts and related background pages for providers and deployers. (digital-strategy.ec.europa.eu) (ai-act-service-desk.ec.europa.eu)