On 7 May 2026, the Council of the European Union and the European Parliament reached a provisional political agreement on the Digital Omnibus on AI, a legislative package postponing by approximately 16 months the application of the strictest obligations under Regulation (EU) 2024/1689 on artificial intelligence.
The key deadline for some high-risk AI systems has been postponed from 2 August 2026 to 2 December 2027, while others are postponed to 2 August 2028.
-
▸
Postponement: Annex III → 2 December 2027; Annex I → 2 August 2028.
-
▸
Fixed deadlines: the conditional mechanism initially proposed has been removed. The dates no longer depend on the completion of harmonized standards.
-
▸
Immediately relevant: transparency obligations for AI-generated content (Article 50) will apply from 2 December 2026.
-
▸
Unchanged: prohibited AI practices (Article 5), AI literacy obligations (Article 4), and obligations for GPAI providers remain applicable under the original timeline.
1. What has changed
The political agreement reached on 7 May 2026 is the result of the Digital Omnibus on AI, proposed by the European Commission on 19 November 2025 as part of a broader EU digital simplification agenda.
Negotiations were difficult. The trilogue session held at the end of April had failed. Still, the final compromise put forward by the Cypriot EU Presidency confirms the postponement in a predictable form, with fixed calendar dates rather than implementation linked to the availability of harmonised standards.
The official rationale for the delay is the incomplete development of harmonised technical standards by CEN and CENELEC, which were mandated by the Commission as early as May 2023. Without harmonised standards, operators cannot benefit from the presumption of conformity; without sector-specific guidance from the Commission, national authorities cannot effectively supervise compliance.
In addition, competent national authorities, including those in Romania, have not yet been designated or fully operationalised.
2. Revised timeline
| Obligation | Original deadline | New deadline |
|---|---|---|
| High-risk AI systems – Annex III | 2 August 2026 | 2 December 2027 |
| High-risk AI systems – Annex I | 2 August 2027 | 2 August 2028 |
| Transparency obligations for AI-generated content | 2 August 2026 | 2 December 2026 |
| National regulatory sandboxes | 2 August 2026 | 2 August 2027 |
| Prohibited AI practices (Article 5) and AI literacy obligations (Article 4) | 2 February 2025 | Unchanged – already applicable |
| Obligations for GPAI providers | 2 August 2025 | Unchanged – already applicable |
3. What remains urgent
Although the postponement provides significant additional time for high-risk systems, three categories of obligations remain on their original or near-original implementation path and require immediate attention:
Transparency for synthetic content (Article 50). The application date has been set for 2 December 2026, less than seven months away.
Operators generating or publishing AI-generated images, video, audio, or text must implement watermarking solutions and inform users accordingly.
This is particularly relevant for media, advertising, e-commerce, publishing, and content marketing sectors.
Prohibited practices (Article 5) and AI literacy obligations (Article 4). Applicable since 2 February 2025, these obligations have not been postponed. The political agreement also introduces a new prohibition covering AI systems generating non-consensual intimate imagery and child sexual abuse material (CSAM).
Obligations for providers of general-purpose AI models (GPAI). Applicable since 2 August 2025 and unchanged. These obligations indirectly affect any operator integrating models such as GPT, Claude, Gemini, Mistral, or Llama into operational workflows through contractual clauses and technical documentation provided by model providers.
4. Practical implications
Our recommendation is clear: the postponement should not be interpreted as a suspension. Operators assuming that deadlines will be postponed again risk losing valuable implementation time. The additional 16 months should be used strategically, not wasted.
- Resume paused compliance projects.
- Reassess system classifications.
- Confirm roles (provider vs. deployer/user).
- Prepare conformity assessments.
- Monitor harmonised standards.
- Update contractual frameworks.
5. What comes next
The agreement reached on 7 May 2026 remains provisional. A technical meeting will follow to finalise the legislative wording, after which formal adoption by the European Parliament and the Council is expected, followed by publication in the Official Journal of the European Union, estimated within approximately two months.
Until publication, the original implementation timeline under Regulation (EU) 2024/1689 formally remains in force.
The ONV LAW team advises companies on AI Act implementation through:
- ▸AI system classification and gap assessments;
- ▸drafting internal AI governance policies;
- ▸reviewing contracts with AI solution providers;
- ▸assistance in dealings with competent authorities; and
- ▸training for management and operational teams.