The EU AI Act’s August 2 date is real — but not the part you were told.

For a year, “August 2, 2026” was shorthand for “high-risk AI obligations arrive.” That is no longer accurate. The date stands; the high-risk wall moved.

What changed in 2026

In May 2026, EU negotiators reached a provisional agreement on the Digital Omnibus on AI — a package that grants timeline relief on the most burdensome obligations. The headline shift: stand-alone high-risk systems under Annex III (recruitment, credit scoring, education, law enforcement, border control) would move to December 2, 2027, and AI embedded in regulated products under Annex I (medical devices, machinery, vehicles) to August 2, 2028. The European Parliament approved the agreed text on June 16, 2026, but these changes take legal effect only on formal Council adoption and publication in the Official Journal — which, as of late June 2026, has not yet happened. Until it does, the original August 2, 2026 high-risk date remains the law (see the dated update below).

What still applies on August 2, 2026

This is the part many companies miss. August 2, 2026 remains an active compliance date for several things that may well touch you:

• Article 50 transparency obligations. If you operate a chatbot, generate synthetic media, or use emotion-recognition or deepfake-capable tools, you owe users clear disclosure. This is the obligation most U.S. companies actually trip on — and as of June 10, 2026 there is now a concrete playbook for it: the European Commission published a (voluntary) Code of Practice on Transparency of AI-Generated Content, setting out practical mechanisms — digitally-signed metadata, imperceptible watermarking, and content labelling — for meeting the Article 50 marking-and-labelling duties that apply from August 2, 2026.
• Governance, penalties, and notified-body machinery. The enforcement architecture switches on.
• General-purpose AI model obligations continue to phase in for providers of foundation models.

Update — June 10, 2026: The EU AI Office published the Code of Practice on Transparency of AI-Generated Content. It is voluntary, but it is the clearest signal yet that the August 2, 2026 marking-and-labelling duties are operational, not theoretical — and signing it is a way to demonstrate good-faith compliance. If your product generates synthetic audio, image, video, or text that reaches the EU, the runway to stand up watermarking/metadata and labelling controls is now measured against that date. We help you scope which Article 50 duties actually apply and build the controls to meet them.

Update — June 29, 2026: Watch the gap between the headlines and the law. The Digital Omnibus that defers the high-risk deadlines was approved by the European Parliament on June 16, 2026, but it still needs formal Council adoption, signature, and publication in the Official Journal — none of which has occurred as of this update. Until the Omnibus is published, the original AI Act timeline is still binding, including the August 2, 2026 applicability date for high-risk systems and the Article 50 transparency duties that start the same day. Our read: treat the deferral as near-certain but not yet law, and keep your August 2 readiness on track until the Official Journal says otherwise. Standing down a readiness program on a “it’s delayed” headline is the avoidable risk here.

Does the EU AI Act even reach a US company?

Often, yes — the Act is extraterritorial. If your AI system’s output is used in the EU, or you place an AI product on the EU market, you can be in scope regardless of where you are headquartered. The honest answer for most mid-sized U.S. firms is “partially”: a customer-facing chatbot or a hiring tool used on EU candidates can pull you into specific obligations without making your whole company a high-risk operator.

The deferral is good news only for the prepared. The companies that benefit are the ones who used the extra sixteen months to build a real management system — not the ones who heard “delayed” and closed the file.

This briefing is general information from Sentinel Assurance Group, not legal advice. Regulatory dates and requirements change — we maintain these briefings, but verify against primary sources and counsel before acting. Last reviewed June 5, 2026.

See our ISO/IEC 42001 readiness path →