01
Source change
On 19 November 2025 the European Commission proposed the digital omnibus on AI (COM(2025) 836), targeted amendments to the AI Act's implementation timing. On 7 May 2026 the European Parliament and the Council reached a provisional agreement to defer the AI Act's high-risk deadlines and postpone part of the content-marking duty — endorsed by Member States' ambassadors on 13 May 2026. Formal adoption and Official Journal publication were still pending when this case was last checked (7 Jul 2026).
Proposed change; until formal adoption the published text applies.
02
Obligation affected
The deferral touches three tracked obligations — the transparency content-marking duty and both high-risk tracks. Each keeps its formally applicable date until the amending regulation is adopted; see the full entries, sources and changelogs on
AI Act deadlines.
Transparency duties under Article 50 (disclosing AI interaction, marking synthetic content, labelling deepfakes) formally apply from 2 August 2026; a provisional agreement would postpone the content-marking duty to 2 December 2026 for systems placed on the market before 2 August 2026.
High-risk obligations for stand-alone AI systems (Annex III — risk management, data governance, documentation, human oversight, conformity assessment) formally apply from 2 August 2026; a provisional agreement would defer this to 2 December 2027.
High-risk obligations for AI embedded in regulated products (Annex I) formally apply from 2 August 2026; a provisional agreement would defer this to 2 August 2028.
03
Impact
Illustrative scenario
Take a deployer running a CV screening model for recruitment — automated decisioning, classified high-risk under Annex III (Art. 6). Under the AI Act as published, this system's high-risk obligations (risk management, data governance, documentation, human oversight, conformity assessment) become applicable on 2 August 2026. Under the provisional agreement, that date would move to 2 December 2027 for this Annex III system — sixteen months of extra runway, but not yet law.
04
Action
What this deployer does now, factually:
- Keep the non-high-risk parts of the AI Act programme on the 2 August 2026 date — the deferral covers the high-risk requirements, not the whole Act.
- Confirm the CV screening model sits in the Annex III track (December 2027 under the agreement) rather than the embedded-product track (August 2028).
- Check whether any generative components placed on the market before 2 August 2026 rely on the content-marking postponement to 2 December 2026.
- Watch the Official Journal for the adopted amending regulation, and take the final date from that text, not from the agreement.
- Record, for audit purposes, which planning decisions were taken on the provisional dates and when.
05
Evidence
Each of those actions leaves a dated entry against the CV screening system in a monitored AI register: the obligation tracked, its formally applicable date, the provisional date under the agreement with its source, and the timestamp of the last verification against the primary text. That per-system record — not a screenshot of any internal tooling — is what a compliance lead hands to counsel, an auditor or a market-surveillance authority: a traceable answer to ‘what did we know, and when, about this deadline.’
Case published: 7 Jul 2026. Reflects the situation as of that date. Not legal advice: Trusq indicates likely attention points; a competent authority or court always has the final say. Proposed change; until formal adoption the published text applies.
Continue
← From EU change to action