The EU Just Delayed Its AI Act. Here's What Still Lands on August 2

If you'd bookmarked August 2, 2026 as the day the EU AI Act's high-risk rules kicked in, you can mostly unbookmark it. On June 16, the European Parliament formally adopted the Commission's Digital Omnibus on AI, pushing most of that compliance burden back by well over a year.

What didn't move is smaller, but it's the part most developers will actually touch.

What got delayed

The Commission first proposed this deferral back in November 2025, arguing that conformity assessment infrastructure, harmonized standards, and notified bodies simply weren't ready in time for the original schedule. Parliament agreed, and the amendment is now law.

If you're building an HR screening tool, a credit-underwriting model, or anything else that falls under Annex III, you effectively got 17 more months. That's real breathing room for conformity assessments, risk management documentation, and human oversight design that most teams hadn't started.

What didn't move

Two things are still landing on schedule, and they're the ones with the shortest runway left:

• Article 50 transparency duties. Disclosing to users that they're interacting with an AI system, labeling AI-generated content, and marking deepfakes. These obligations were explicitly untouched by the omnibus and take effect August 2, 2026, three weeks from now.
• General-purpose AI model enforcement. The Commission's enforcement powers over GPAI models activate the same day, meaning the obligations that already applied to foundation model providers now have teeth behind them.

[!IMPORTANT] If your product includes a chatbot, a synthetic voice, or any AI-generated image or video shown to EU users, the disclosure and labeling requirements apply regardless of whether your system counts as "high-risk." That distinction is what got delayed, not this one.

Why this matters even if you're a US company

Several US-focused compliance write-ups this month have flagged the same thing: American companies serving EU users don't get to sit this out just because they're not headquartered in Brussels. The AI Act applies based on where the system is used, not where the vendor is based. A US SaaS product with EU customers and an AI chatbot needs Article 50 disclosures live by August 2, regardless of the high-risk delay.

The practical checklist

For most teams building AI-facing products, the next three weeks matter more than the next 17 months:

1. Audit for user-facing AI interactions. Chatbots, voice assistants, and generative content need clear disclosure that the user is talking to or viewing AI output.
2. Add content labeling for anything AI-generated that ships to end users, images, video, and audio in particular.
3. Don't stop tracking your Annex III exposure just because the deadline moved. Seventeen months disappears fast, and conformity assessment processes for genuinely high-risk systems (hiring, credit, biometrics) take real time to build.

The broader signal here is that EU AI regulation isn't retreating, it's sequencing. Transparency first, because it's cheap to implement and hard to argue against. High-risk conformity assessment later, because the infrastructure to actually assess it wasn't ready. Don't read the delay as the EU backing off. Read it as the EU admitting its own bureaucracy needed more time to catch up to its own law.