This page is a summary, not legal advice.
The EU AI Act sets obligations according to the use of an AI system and the organisation's role. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was published on 24 July 2026 and entered into force on 27 July 2026. Compliance planning should use the amended law.
Which deadlines apply?
The main high-risk deadlines are now 2 December 2027 for Annex III systems and 2 August 2028 for regulated-product systems under Annex I. The transparency rules have a different timetable.
| Obligation | Provision | Application date | Primary source | Verification |
|---|---|---|---|---|
| Original prohibited practices | Article 5 | 2 February 2025 | AI Act | Last verified |
| Measures supporting staff AI literacy | Article 4, as amended | In application since 2 February 2025; amended wording from 27 July 2026 | Commission Q&A | Last verified |
| General-purpose AI model obligations | Chapter V | 2 August 2025; transitional provisions apply to older models | Commission overview | Last verified |
| Transparency, including chatbot disclosure and deepfake disclosure | Article 50 | 2 August 2026 | Implementation timeline | Last verified |
| Machine-readable marking of synthetic outputs from systems placed on the market before 2 August 2026 | Articles 50(2) and 111(4) | Transition ends 2 December 2026 | Omnibus, official text | Last verified |
| New prohibitions concerning non-consensual intimate material and child sexual abuse material | Article 5, as amended | 2 December 2026 | Omnibus, application provisions | Last verified |
| Stand-alone high-risk systems, subject to the legal classification and exceptions | Article 6(2), Annex III | 2 December 2027 | Council timeline | Last verified |
| High-risk AI embedded in regulated products | Article 6(1), Annex I | 2 August 2028 | Council timeline | Last verified |
The marking transition is specific to Article 50(2) and qualifying systems already on the market. It does not postpone all of Article 50. The official text also contains transitional rules for older systems and models; check those against the actual product and market-entry date.
What changed in AI literacy?
Providers and deployers must take measures that support the development of AI literacy among staff and people operating AI on their behalf. The amended Article 4 does not require a particular or “sufficient” level for each person.
The Commission's AI literacy Q&A explains how measures can reflect people's responsibilities, technical background and context of use. Record the tools, users, risks and practical support covered. A course attendance certificate alone does not establish compliance with the rest of the Act.
A leadership team can use a private executive briefing to identify the decisions it needs to resolve. Obtain legal advice for classification and compliance conclusions.
Which systems need a closer review?
Start with the actual purpose of each system, then check the relevant legal definition and exceptions. A generic label such as “HR AI” is not a complete legal classification.
Annex III includes specified uses in areas such as employment, education, biometrics and access to essential services. Annex I concerns AI connected with regulated products. Providers and deployers have different responsibilities: identify the party placing a system on the market, the organisation using it, and any changes to the original intended purpose.
Review the Spanish oversight context and the enterprise adoption guide alongside this inventory. The Summit's 2026 programme offers practitioner perspectives; it does not replace legal assessment.
What do the new prohibitions cover?
The Omnibus adds prohibited uses involving AI-generated or manipulated non-consensual intimate material and child sexual abuse material. Their application date is 2 December 2026, under the amending regulation's specific application provisions. Read the operative wording and any qualifications in the official text before translating these descriptions into product controls.
What should a company do next?
Create an inventory of systems, purposes, users, vendors and deployment dates. For each entry, assign a review owner, identify the legal role, check prohibited practices, assess high-risk classification and examine transparency duties.
Keep evidence of decisions, human oversight and staff support. Prioritise obligations already in application while preparing for the later high-risk deadlines. Executive readiness can help structure the internal discussion; a named legal adviser should review the conclusions.
Sources
Frequently asked questions
Is the Digital Omnibus on AI adopted law?
Yes. Regulation (EU) 2026/1744 was published on 24 July 2026 and entered into force on 27 July 2026.
Does the AI literacy obligation require a certification?
The amended Article 4 requires measures supporting AI literacy, without a specific required level for each person. The Commission does not prescribe a universal certification.
Has all synthetic-content transparency been postponed?
No. Article 50 applies from 2 August 2026. A specific transition until 2 December 2026 concerns Article 50(2) marking for qualifying systems placed on the market before 2 August 2026.
Change log
- — Replaced the obsolete Omnibus proposal timeline with the adopted regulation, revised Article 4, Article 50 transition and new prohibited-practice dates. Added primary sources and legal-summary notice.
Written by
- Guillaume RostandCEO, AI Summit Barcelona
Guillaume leads AI Summit Barcelona. His editorial work covers AI events and the European AI ecosystem.
- Tanguy WinckerCo-founder, AI Summit Barcelona
Tanguy is a co-founder of AI Summit Barcelona. His editorial work covers company programmes, event planning and the Barcelona ecosystem.
- Adam HruskaCo-founder, AI Summit Barcelona
Adam is a co-founder of AI Summit Barcelona and contributes to its guides on AI events and the European ecosystem.
Related reading
Enterprise AI Adoption: From Awareness to Useful Experiments
A decision guide for companies moving beyond AI awareness. Diagnose the barrier, choose the appropriate format and define the evidence needed before expanding an experiment.
AESIA: Spain's AI Supervisory Agency
Understand AESIA, Spain’s AI oversight framework and the status of the national governance bill, with links to official legislative sources.
Large language models, compared
Explore differences between models and the trade-offs that deserve investigation. A comparison can help you prepare questions about capabilities and limitations. Before selecting a tool, connect those questions to the task, information and review process that matter in your own working environment.