The deadline moved. The obligations didn't disappear.
On June 29, 2026 the EU finalized the Digital Omnibus on AI: high-risk obligations are deferred to December 2027 and August 2028. Much of the internet is now out of date — and much of it also misses what still happens on August 2, 2026: Article 50 transparency obligations become enforceable, together with the Act's penalty regime.
What applies when — after the 2026 Omnibus.
As amended by the AI Act simplification package (Parliament: June 16, 2026 · Council: June 29, 2026).
| Date | What applies | Status |
|---|---|---|
| Feb 2, 2025 | Prohibited AI practices ban; AI literacy duties | In force |
| Aug 2, 2025 | General-purpose AI model obligations; governance bodies | In force |
| Aug 2, 2026 | Article 50 transparency obligations become enforceable; national market-surveillance authorities and penalties apply | Weeks away |
| Dec 2, 2027 | High-risk obligations for stand-alone Annex III systems (recruitment, credit scoring, education, law enforcement…) | Deferred from Aug 2026 |
| Aug 2, 2028 | High-risk obligations for AI embedded in regulated products (Annex I) | Deferred from Aug 2027 |
The Article 50 readiness checklist.
If any of these apply to a system you provide or deploy in the EU, you have obligations in a matter of weeks. Fines for transparency violations reach €15M or 3% of worldwide turnover (Art. 99(4)).
- Chatbots & conversational AI — people must be told they are interacting with a machine
- Synthetic audio, image, video or text — machine-readable marking of AI-generated content
- Deepfakes — clear disclosure that content is artificially generated or manipulated
- Emotion recognition & biometric categorisation — affected persons must be informed
- AI-written text published on matters of public interest — disclosure required
- Evidence — you must be able to prove each disclosure existed, per system, per version, on a given date
rule EUAIA_ART50_SCOPE
triggers_if:
- interacts_with_natural_persons
- generates_synthetic_content
- performs_emotion_recognition
obligations: disclose, mark, label
evidence: disclosure_artifact, timestamp,
signed_hash # admissible, reproducible
Why "wait until 2027" is the wrong conclusion.
Rules already in force stay in force
The prohibited-practices ban and AI-literacy duties (Feb 2025) and general-purpose AI obligations (Aug 2025) were never deferred. Enforcement infrastructure goes live August 2, 2026.
Procurement doesn't read statutes
Enterprise buyers and insurers increasingly require ISO/IEC 42001-aligned governance evidence in RFPs and underwriting today — independent of any statutory date.
December 2027 is closer than it looks
Conformity assessment for high-risk systems typically takes 12–18 months of inventory, documentation, testing and evidence work. Organizations that start in 2027 will not finish in 2027.
78% of organizations aren't ready anyway
Industry surveys in 2026 found 78% of organizations had taken no meaningful compliance steps and over half lacked a basic AI inventory — the first artifact every framework requires.
Straight answers.
Did the EU AI Act deadlines really move?
What exactly happens on August 2, 2026?
How does AIOPS handle Article 50?
Is this legal advice?
Be ready before the enforcement machinery switches on.
Inventory your systems, detect Article 50 scope, and generate signed disclosure evidence — before August 2, not after the first authority letter.