Over 10 years we help companies reach their financial and branding goals. Engitech is a values-driven technology agency dedicated.

Gallery

Contacts

411 University St, Seattle, USA

engitech@oceanthemes.net

+1 -800-456-478-23

Skip to content
Regulation Deep Dive

Who Wins, Who Waits, and Who Must Still Comply: EU AI Act Delay Proposal Mapped by Stakeholder

The Digital Omnibus on AI reportedly proposes delaying certain EU AI Act high-risk system obligations by 16 months or more, welcome news for some operators, irrelevant noise for others, and a potential trap for anyone who treats a proposed delay as an adopted one. The negotiations simultaneously expand Article 5 prohibitions, meaning the same trilogue that offers a deadline extension in one direction tightens requirements in another. Understanding which category your organization falls into is the compliance decision that matters right now.

A proposed 16-month delay on EU AI Act high-risk system compliance sounds like the same thing for every affected company. It isn’t. Where you sit in the EU AI Act’s compliance architecture determines whether the proposal is relief, distraction, or a trap, and the answer is different for GPAI model providers, Annex III system operators, companies with February 2025 and August 2024 obligations already in effect, and organizations building systems in the content categories that Article 5 is now expanding to cover.

This analysis maps the proposed delay against each major stakeholder category. The goal is to answer the question the daily brief raises but can’t fully address: given the proposed timeline shift, what should different types of organizations actually do right now?

Where the Proposal Stands

The Digital Omnibus on AI (COM(2025) 836) is in trilogue, the negotiation between the European Parliament, Council, and Commission. According to practitioner tracking of Council trilogue positions, the Council has reportedly proposed delaying certain Annex III high-risk obligations to as late as December 2027 (some categories) or August 2028 (others), contingent on harmonized standards availability. Official Council text should be consulted for precise scope, these dates reflect practitioner analysis of Council positions, not adopted language.

What’s confirmed: the EU AI Act’s general application date is August 2, 2026. That is established public record. It is the operative deadline until the Digital Omnibus is formally adopted through the full EU legislative process and published in the Official Journal. That hasn’t happened. The delay is a proposal.

The Article 5 expansion is also confirmed, not as adopted law, but as a subject of ongoing trilogue negotiations with T1 source support. Official EU documentation indicates the negotiations include a proposed new prohibition covering AI systems that generate non-consensual sexually explicit content. This moves in the opposite direction from the timeline delay: one negotiating stream proposes giving operators more time; the other proposes adding new categories of prohibited conduct.

Stakeholder Category 1: Operators Already Under Earlier Obligations

The proposed Annex III delay is specifically about the high-risk system provisions that apply from August 2, 2026. It says nothing about earlier provisions already in force.

The EU AI Act’s prohibited practice provisions applied from February 2, 2025. GPAI provisions and governance framework requirements applied from August 2, 2025. Those obligations are not subject to the proposed delay, they’re already law, already operative, and already being monitored by member state authorities.

For organizations whose primary compliance work involves these earlier phases, the Digital Omnibus delay is not relevant to their current obligations. It may affect planning for August 2026 additions to existing programs, but it doesn’t change what’s already required of them today.

Stakeholder Category 2: Annex III System Operators

This is the group for whom the proposed delay is most directly relevant, and also the group most at risk of misreading it.

Annex III of the EU AI Act identifies eight high-risk application categories: biometric identification systems, critical infrastructure management, educational systems, employment-related AI, essential service access systems, law enforcement applications, border control systems, and justice administration. Operators building or deploying systems in these categories were planning for an August 2, 2026 compliance deadline.

The proposed Council position would push some of these obligations to December 2027 or August 2028. That’s potentially 16 to 24 additional months. For organizations that are genuinely behind, those that haven’t started conformity assessments, haven’t mapped their technical documentation requirements, haven’t established quality management systems, the delay could matter operationally.

But three risks apply to any operator considering a compliance pause.

Risk 1: The delay doesn’t pass. Trilogue negotiations fail, stall, or produce outcomes that look nothing like initial positions. Parliament has historically been more protective of the Act’s enforcement architecture than the Council. If the delay proposal doesn’t survive trilogue, operators who paused their programs face a compressed sprint back to August 2026 readiness.

Risk 2: The delay is conditional. Per practitioner analysis, the proposed Council position ties the delay to harmonized standards availability. If harmonized standards are published before the proposed extended deadline, the delay may not apply to all Annex III categories. The “deadline” in the delay proposal may itself be a moving target.

Risk 3: Compliance work is not wasted even if the delay passes. Conformity assessments, technical documentation, and quality management systems built for August 2026 don’t become worthless if the deadline extends. They’re the foundation of an ongoing compliance program. Organizations that have done this work are better positioned in 2027 or 2028 than those that waited.

Stakeholder Category 3: GPAI Model Providers

Current practitioner interpretation holds that the voluntary GPAI Code of Practice will remain the primary compliance mechanism through the GPAI obligation phase-in period, per the EU AI Act compliance tracker. The Digital Omnibus could affect these dates, and GPAI providers should monitor for official Commission guidance rather than treating the Code of Practice timeline as fixed.

For GPAI providers, the more immediately relevant question is whether the Article 5 expansion affects their models’ output capabilities. A model that can generate non-consensual explicit content, even if that’s not the intended use case, may fall within the scope of the proposed new prohibition regardless of what happens to the Annex III timeline. The proposed Article 5 prohibition is about content categories, not about when compliance is required. It’s a harder constraint than a deadline extension can soften.

Stakeholder Category 4: Organizations in Article 5 Categories

The proposed Article 5 addition covers AI systems that generate non-consensual sexually explicit content. If your organization builds, operates, or enables systems in this category, the Digital Omnibus is not bringing relief, it’s bringing additional exposure.

Article 5 prohibitions have applied from February 2, 2025. The proposed expansion adds a new category to that existing prohibition structure. If adopted, it wouldn’t come with the kind of phase-in timeline that Annex III obligations carry. Prohibitions under Article 5 are the EU AI Act’s hardest requirements.

Organizations in adjacent categories, synthetic media generation platforms, content moderation AI, consumer-facing generation tools, should be actively monitoring what the final Article 5 text includes and what defenses the regulation provides. The proposed text language, as confirmed through official EU sources, targets non-consensual content specifically. Systems with robust consent and authorization frameworks may be positioned differently than systems with permissive generation defaults.

What Compliance Teams Should Do Right Now

The decision hierarchy is straightforward even if the regulatory picture isn’t.

Continue August 2026 preparations. The delay is proposed, not adopted. A completed conformity assessment is a durable asset; a paused one is a liability if the delay fails.

Map your Article 5 exposure specifically. The same trilogue that may extend your Annex III deadline is adding new prohibitions. Know which of those apply to your systems before the final text is adopted, not after.

Treat the GPAI Code of Practice as a real compliance commitment. It’s voluntary in name; it’s the only compliance mechanism GPAI providers have right now. Participation and documentation matter for demonstrating good faith if enforcement attention comes earlier than expected.

Watch Parliament’s position. The Council’s proposed delay reflects competitiveness concerns. The Parliament has historically prioritized enforcement integrity. The final trilogue outcome will reflect that tension. Parliament’s posture is the leading indicator of whether the delay survives.

TJS Synthesis

The Digital Omnibus delay proposal tells compliance teams two things simultaneously: some obligations may come later, and some prohibitions will be broader than previously planned. Those two signals pull in opposite directions, and they don’t apply to the same stakeholder categories. The organizations best positioned to handle this uncertainty are the ones that built August 2026 compliance programs regardless, they have the foundation to adapt to either outcome. The ones most exposed are those that read “proposed delay” as “confirmed extension” and treated a practitioner’s Council position summary as a compliance decision. Proposed is not adopted. August 2, 2026 still stands.

View Source
More Regulation intelligence
View all Regulation
Related Coverage

More from April 23, 2026

Stay ahead on Regulation

Get verified AI intelligence delivered daily. No hype, no speculation, just what matters.

Explore the AI News Hub