The Omnibus delayed the EU AI Act. Twice, it didn’t.

The headline after the EU’s Digital Omnibus was “high-risk AI delayed to 2027.” True — and misleading. Two of the things a U.S. company is most likely to trip over were not delayed at all, and one of them lands the same day as a California law. Here is the honest version of what changed and what still bites.

What the Omnibus delayed

The Omnibus gave real breathing room on high-risk AI. Annex III stand-alone high-risk obligations moved from August 2, 2026 to December 2, 2027 — a sixteen-month deferral. High-risk AI embedded in regulated products (Annex I) moved to August 2, 2028. If your system is a high-risk classifier — credit scoring, hiring, biometric categorization — you have more time to reach conformity than the original timeline gave you.

What it did NOT delay (twice)

Here is what the “everything got delayed” reading gets wrong. First, Article 50 transparency obligations — telling users they’re interacting with AI, and labeling AI-generated content — still take effect August 2, 2026 (watermarking of pre-existing content gets a short grace to December 2, 2026). Second, general-purpose AI (GPAI) model obligations remain in force from August 2, 2025, and the Act’s prohibited practices have applied since February 2, 2025. So “everything got delayed” is wrong twice over — and for a U.S. company, Article 50 and California’s Transparency Act both land on August 2, 2026.

Does it even reach a U.S. company?

Often, yes. Like the GDPR, the AI Act is extraterritorial: it applies if you place an AI system on the EU market, or if the output of your AI is used in the EU — regardless of where you sit. If you have EU users, EU customers, or an EU entity, you should assume Article 50 reaches your customer-facing chatbots and any AI-generated content you publish into the EU. The threshold question isn’t “are we a European company?” It’s “does our AI touch anyone in Europe?”

The transatlantic pairing

August 2, 2026 is a double date. On the same day, the EU’s Article 50 and California’s AI Transparency Act (CAITA) both take effect — and both are fundamentally about disclosure and provenance: tell people when content or an interaction is AI, and be able to prove it. California’s SB 1001 has required bot disclosure since 2019. The efficient move is to build one AI-interaction-disclosure and content-labeling capability that satisfies all three at once, rather than three separate projects under three regimes.

A note on status — and on being honest about it

The Digital Omnibus was politically agreed in late 2025 and adopted in mid-2026, entering force ahead of the August 2 milestone. As with any fast-moving instrument, verify the final Official Journal text before you rely on precise wording — but the direction is settled, and the near-term dates above are the ones to plan around. This is the honest posture we take with clients operating across the U.S. and EU: the high-risk clock slowed, the transparency clock did not.

The Omnibus is good news and a trap at the same time: it bought time on the hardest obligations while leaving the nearest one exactly where it was. For a U.S. company, the date that matters is still August 2, 2026 — on both sides of the Atlantic.

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 July 27, 2026.