June 23 matters. Not as a compliance deadline, your systems don’t need to be ready by then. But as a participation deadline. The Commission published draft guidelines on high-risk AI classification criteria under EU AI Act Article 6 on May 19, 2026. Legal analysis from Gibson Dunn and Faegre Drinker published in the May 27-29 window frames the consultation window as the last formal mechanism for organizations to influence how Annex III categories get defined before those definitions carry legal weight.
Here’s what the consultation covers. The draft guidelines clarify how the Commission interprets the “significant risk of harm” exception under Article 6(1)(a), the provision that allows certain AI systems to avoid high-risk classification even when they fall within an Annex III category. For organizations deploying AI in employment screening, credit scoring, biometric identification, or educational assessment, the exception’s precise scope determines whether their systems face the full conformity assessment process or a lighter-touch pathway. These categories aren’t hypothetical. They describe deployed systems at thousands of organizations operating in the EU market today.
The compliance timeline has a structure worth stating plainly. Stand-alone Annex III high-risk systems face a December 2, 2027 compliance deadline under the Digital Omnibus provisional agreement. Embedded product-safety systems under Annex I face an August 2, 2028 deadline. Both of those deadlines feel distant. The consultation window is not distant. It closes in 25 days and won’t reopen.
Who This Affects
Article 50 transparency obligations run on a different track. Legal analysis discussed in prior TJS coverage has drawn a distinction between two sub-obligations: Article 50(1) chatbot and AI-interaction disclosure requirements, which are reported as active from August 2, 2026, and Article 50(2) synthetic content labeling obligations, which carry a December 2026 effective date according to legal analysis attributed to Gibson Dunn. [HUMAN LEGAL REVIEW REQUIRED: The August vs. December distinction for Article 50 sub-obligations carries compliance risk if misstated. Do not publish this paragraph without legal verification of the Article 50(1)/50(2) split.] Organizations deploying chatbots or AI-generated synthetic content should verify directly with EU AI Act counsel which obligation applies to their deployment before August 2026.
What to watch
the Commission will process consultation submissions and may revise the draft guidelines before finalizing the high-risk classification criteria. Organizations in Annex III categories that submit well-reasoned comments can influence whether their specific use case falls inside or outside the “significant risk of harm” exception. Law firm analysis from Faegre Drinker notes that the draft guidelines’ treatment of the exception is where the practical compliance stakes are highest.
June 23 Consultation Preparation, Minimum Steps
- Map EU-deployed AI systems against Annex III categories
- Identify systems potentially qualifying for Article 6(1)(a) 'significant risk of harm' exception
- Review draft guidelines for classification criteria relevant to your sector
- Engage EU AI Act counsel for Article 50(1)/50(2) applicability determination
- Draft and submit consultation comments before June 23
Don’t expect a light workload for a consultation submission. A useful submission requires the organization to map its specific AI deployment against the proposed classification criteria, identify where the draft guidelines are ambiguous or overbroad relative to that deployment, and make a reasoned argument for how the Commission should interpret the exception. That’s a legal and technical exercise. Twenty-five days is enough time to complete it with the right team engaged, but not if the deadline arrives before anyone starts.
The real question is how many compliance programs have prioritized this. The December 2027 compliance deadline gets the attention. The June 23 consultation window is the action that shapes what December 2027 actually requires. Organizations that submit credible technical comments have a direct line to influencing the final classification criteria. Those that don’t, accept the Commission’s draft as written.