What Changed on June 10, and What Changed in May
The provisional agreement on the EU AI Act’s Digital Omnibus amendments was reached on May 27, 2026. The hub covered it then. What’s new on June 10 isn’t the agreement, it’s what came out of it: the European Commission’s release of draft non-binding guidelines on high-risk AI system (HRAI) classification, and the activation of a public consultation window that closes June 23. These are two distinct deliverables. The agreement set the deadlines. The guidelines are the first attempt to clarify what those deadlines actually apply to.
That distinction matters for compliance planning. An organization can note a deadline extension and move on. It can’t ignore draft classification guidelines that will determine whether its systems are subject to that deadline at all. The June 23 consultation is where the scope of “high-risk” gets contested before it gets fixed. According to the EU AI Act’s official tracker, Article 50 transparency rules and the August 2026 baseline application date remain in force, the Omnibus amendments layer on top of them, not in place of them.
The Deadline Map by Organization Type
Four deadlines are now in play, and which one governs your organization depends on how your AI systems are classified. Here’s the complete picture, based on analysis from Gibson Dunn and DLA Piper:
| Organization Type | Relevant Annex | Revised Deadline | Prior Deadline | Status |
|---|---|---|---|---|
| Use-based HRAI deployers (employment, education, critical infrastructure, etc.) | Annex III | December 2, 2027 | August 2, 2026 | Provisional, pending Official Journal |
| AI embedded in sector-regulated products (medical devices, machinery, vehicles) | Annex I | August 2, 2028 | August 2, 2026 | Provisional, pending Official Journal |
| Pre-existing generative AI systems (on market before Aug 2, 2026) | Article 50 | December 2, 2026 | August 2, 2026 | Provisional, pending Official Journal |
| GPAI model providers | Chapter V | August 2, 2026 (unchanged) | August 2, 2026 | No change, in force |
The Annex III deferral is 16 months. Gibson Dunn confirms the December 2, 2027 date and the arithmetic directly. Fisher Phillips, analyzing the same agreement, frames it practically: “compliance deadline for high-risk AI systems used in employment decisions just moved from August 2026 to December 2027.” That’s the headline number, but it’s also the number most likely to generate false comfort. Sixteen months sounds substantial. Compliance programs for Annex III systems require risk assessments, technical documentation, conformity assessments, and post-market monitoring structures. Organizations that treat December 2027 as a start date, rather than a finish line, are already behind.
The Annex I deferral to August 2028 applies to AI embedded in products covered by existing EU product safety legislation, medical devices, general machinery, vehicles. SureCloud’s analysis of the agreement notes this explicitly: “Annex I high-risk systems (product-embedded): Obligations deferred to 2 August 2028. This covers AI components in medical devices, machinery.” The two-year runway is longer, but it interacts with existing conformity assessment cycles for the underlying products. For a medical device manufacturer already in a notified body review process, the August 2028 AI Act deadline doesn’t exist in isolation.
One claim in wider circulation deserves qualification: some sources have described AI deployed in industrial machinery as receiving a “complete exemption” from the Act. The Annex I deferral to August 2028 covers machinery-embedded AI, that’s confirmed. Whether any category of industrial machinery AI receives a full exemption rather than a deferral isn’t independently verifiable from the sources available to this analysis. The scope of any machinery-specific exemption should be confirmed against the official Digital Omnibus text before updating compliance programs.
What the Draft Guidelines Actually Cover
The Commission’s draft non-binding guidelines on HRAI classification address which AI systems fall within the Article 6 and Annex III use-case categories that trigger the full HRAI compliance regime. “Non-binding” is the right framing for the document’s legal weight – but guidance documents from the Commission have historically shaped how national competent authorities interpret and enforce the underlying legislation. That’s why the consultation matters.
Who This Affects
EU AI Act Compliance Steps, Priority Order Before June 23
- Review draft non-binding HRAI classification guidelines (Commission publication)
- Map AI systems against Annex I vs. Annex III criteria to determine governing deadline
- Identify gray-zone systems that fall near Annex III use-case boundaries
- Draft consultation submission with specific system-type examples for Commission review
- Submit before June 23, 2026 consultation close
- Confirm Official Journal publication of Digital Omnibus before updating compliance program timelines
What the hub hasn’t independently verified is the substantive classification criteria within the guidelines themselves, the specific use-case thresholds, carve-outs, or clarifications the Commission has proposed. The consultation response window is partly an opportunity to fill that gap: organizations that review the draft text directly and submit responses with concrete examples will generate better compliance intelligence than those waiting for law firm summaries. The June 23 deadline is firm. Draft guidelines open for comment aren’t typically reopened after the window closes.
The real question is whether your AI systems’ classification as high-risk is clear-cut under Article 6(1) and Annex III, or whether they sit in a gray zone where the draft guidelines’ classification criteria will determine your compliance path. For systems squarely within Annex III, AI used in hiring, credit scoring, education access, law enforcement – the answer is clear and the deadline is December 2027. For systems in adjacent categories, the consultation is where the line gets drawn.
The June 23 Consultation: Who Should Submit and What to Say
Three types of organizations have the most to gain from submitting before June 23.
Deployers with mixed-use systems. AI systems that serve multiple functions – some of which fall clearly within Annex III categories and some of which don’t, face classification ambiguity that the draft guidelines are meant to resolve. A submission that presents specific system architectures and asks for classification clarity is exactly what the consultation process is designed to accommodate.
Developers in Annex I product categories. The interaction between the AI Act’s Annex I product scope and the existing product safety legislation it references – Directive 2006/42/EC on machinery, Regulation 2017/745 on medical devices, creates compliance layering questions that the draft guidelines may or may not address. If they don’t, a consultation submission is the mechanism to flag that gap.
Organizations with GPAI components in use-based systems. The GPAI obligations under Chapter V apply on the current August 2026 timeline, unchanged. But many Annex III deployers are building on top of GPAI foundations. The interaction between GPAI provider obligations and downstream deployer obligations isn’t always clean, and the draft guidelines are an opportunity to push for clarity before December 2027 becomes an active compliance deadline.
The Commission is asking a specific question: do the draft classification criteria accurately capture which systems should carry the full HRAI burden? The most useful submissions are structured around that question, anchored to specific system types, and submitted before June 23 closes the window.
Unanswered Questions
- Does the draft guidelines' classification criteria create a clear threshold for AI systems that serve both Annex III use cases and general-purpose functions?
- What documentation does the Commission expect in a consultation submission, system architecture descriptions, use-case narratives, or specific Article 6(1) analysis?
- How will the interaction between GPAI Chapter V obligations (August 2026) and Annex III deployer obligations (December 2027) be addressed in practice when deployers build on foundation models?
Verification
Partial Cross-reference analysis (Gibson Dunn T2, DLA Piper T2, artificialintelligenceact.eu T1, Fisher Phillips T3, William Fry T3) Industrial machinery 'complete exemption' claim is unverified, available sources confirm an Annex I deferral to August 2028 but do not establish a full exemption. All deadline changes are provisional pending Official Journal publication.The Formal Adoption Gap, What Compliance Teams Can Rely On Now
Gibson Dunn is precise on this point: “These changes only take legal effect upon formal adoption and publication of the Omnibus in the Official Journal, expected before 2 August 2026.” The provisional agreement is politically binding on the EU institutions that reached it. It isn’t legally binding on the organizations subject to the Act until the Official Journal publication occurs.
That creates a specific planning window. Between now and the Official Journal publication – expected before August 2, 2026, compliance teams are operating on provisional terms. The August 2026 deadline for Annex III use-based systems remains technically in force until the Official Journal publishes the deferral. In practice, enforcement against a provisional agreement that has already secured trilogue approval would be extraordinary, but “extraordinary” isn’t “impossible.” The risk profile depends on your jurisdiction, your sector, and your organization’s tolerance for operating on provisional rather than enacted terms.
What compliance teams can rely on now: the classification guidance process is real and the June 23 consultation window is active. The draft guidelines exist regardless of the Omnibus’s formal adoption status. Engaging the consultation, completing HRAI classification assessments, and beginning gap analysis against the December 2027 requirements are all activities with value independent of the formal adoption timeline.
TJS Synthesis
The Digital Omnibus amendments have given compliance teams more time, 16 months more for Annex III systems, but the draft guidelines published June 10 mark the point at which “more time” starts converting into specific compliance requirements. Every month between now and December 2027 that passes without a completed HRAI classification assessment is a month of planning runway consumed. Organizations that treat the June 23 consultation as optional are forfeiting input on the classification criteria that will govern their December 2027 obligations. The classification decision made in the guidelines, high-risk or not – determines whether an organization owes a full conformity assessment, technical documentation package, and post-market monitoring system, or faces a far lighter compliance burden. That boundary is being drawn right now, and it closes in 13 days. The next opportunity to contest it will be in enforcement proceedings, not consultation responses.