European Commission narrows high-risk AI

The European Commission on May 19 published draft guidance on how companies and regulators should decide whether an AI system counts as “high-risk” under the EU AI Act. The documents set out the Commission’s interpretation of Article 6 of the law and add practical examples of systems that should, and should not, fall into the category. The guidance is not legally binding, but the Commission said it is meant to guide enforcement and help providers, deployers and market surveillance authorities apply the rules consistently. ### Which part of the AI Act is the Commission trying to pin down? Article 6 of Regulation (EU) 2024/1689 creates the AI Act’s two main high-risk routes. One covers AI used as a safety component of products, or AI that is itself a product, when that product falls under EU product-safety legislation in Annex I and requires third-party conformity assessment. The other covers stand-alone AI systems used in specific areas listed in Annex III. (digital-strategy.ec.europa.eu) The Commission said the draft is split the same way. One section covers general principles, one covers Annex I product-related systems, and one covers Annex III use cases. The aim, it said, is to help firms work out not only whether a system is high-risk, but which high-risk category it falls into. (digital-strategy.ec.europa.eu) ### What is changing for companies that use AI privately rather than for public functions? The Commission said the draft contains practical examples showing when systems should or should not be classified as high-risk. That matters most in Annex III categories, where the trigger is not the technology alone but the intended use in areas such as employment, education, migration, biometrics or access to essential services. (digital-strategy.ec.europa.eu) The guidance page says only a limited set of use cases are high-risk, specifically those that endanger health, safety or fundamental rights. In practice, that narrows the field for companies that had feared broad categories in the law could sweep in internal or private-sector uses that do not match the listed use cases. The Commission has not presented the examples as exhaustive and said they may be updated over time. (digital-strategy.ec.europa.eu) ### Why does this draft matter now, before the rules fully bite? The AI Act entered into force on Aug. 1, 2024, and set a risk-based framework with separate obligations for prohibited, transparency, high-risk and lower-risk systems. For high-risk AI, the law imposes stricter duties, including compliance, documentation and oversight requirements. (digital-strategy.ec.europa.eu) The timing has shifted. The Commission’s guidance page says that, following the political agreement on the AI Omnibus, rules for stand-alone systems used in high-risk Annex III areas now apply from Dec. 2, 2027, while rules for systems integrated into products such as robotics and industrial machinery apply from Aug. 2, 2028. The Council of the European Union said on May 7 that Council and Parliament negotiators had reached a provisional agreement to simplify and streamline parts of the AI rulebook. (commission.europa.eu) ### What does the Commission say firms should do with the draft now? The Commission opened a targeted consultation on May 19 and said feedback will run until June 23, 2026. Registered participants will receive an online questionnaire, and the Commission said only responses submitted through that process will be reflected in the final summary report. (digital-strategy.ec.europa.eu) The consultation page says the audience includes AI providers, developers, organisations using AI systems, public authorities, researchers, civil society groups, supervisory bodies and members of the public. The Commission has also put the material on its AI Act Single Information Platform in what it calls a more user-friendly format, including summaries and an explorer for examples and use cases. (digital-strategy.ec.europa.eu) ### What comes next before the guidance is final? June 23, 2026, is the deadline for comments on the draft guidelines. The Commission said feedback from that consultation will be incorporated into the final version before formal adoption, while the broader AI Omnibus changes still require final legislative adoption after the May 7 provisional agreement between the Council and Parliament. (digital-strategy.ec.europa.eu)