EU AI Act for High-Risk AI Agents: A Build Checklist
Paused everything when you read “delayed”? The AI Act has applied since 2 August 2026; mainly the high-risk rules moved.
IN 30 SECONDS
The EU AI Act is the European law on AI systems, and the rules in this checklist cover systems it calls high-risk. They start on 2 December 2027 or 2 August 2028, but the Act itself, with Article 72 monitoring, has applied since 2 August 2026. Not legal advice: whether a rule binds your system is a lawyer’s call.
KEY TAKEAWAYS · 3
- The AI Act has applied since 2 August 2026. Regulation (EU) 2026/1744 moved the high-risk requirements to 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
- Article 72, post-market monitoring, kept the 2026 date. What it asks of a given team before 2027 is a question for a lawyer.
- Not legal advice. The build list, which supports compliance work without establishing it: an audit record (Article 12), abstain-and-escalate with a human sign-off (Article 14), a release-gating eval suite (Article 15).
On this page
The AI Act was not delayed
IN SHORT The Act has applied since 2 August 2026. The main change moved the high-risk rules to 2027 and 2028.
WORDS TO KNOW
- EU AI Act
- The European law on AI systems, Regulation (EU) 2024/1689. It entered into force on 1 August 2024 and applies in stages.
- High-risk system
- An AI system the Act classes as high-risk: most Annex III uses, or a product or safety part that Annex I law sends for an outside check.
- Post-market monitoring
- Watching a system once it is in use, and acting on what you find. Article 72 covers it.
That date sits in one sentence of Article 113. The amending law didn’t touch that sentence. What moved is mainly the set of rules for a High-risk system An AI system the Act classes as high-risk: most Annex III uses, or a product or safety part that Annex I law sends for an outside check. See the glossary (Chapter III, Sections 1 to 3).
The headline reading
What Article 113 says
A team that stops building when it reads “delayed” is reading a headline, not Article 113.
For an engineer, that split matters: these rules no longer share one date. This checklist suggests what to build, to discuss with your legal and compliance reviewers.
WATCH OUT
Not legal advice: I’m a smart-contract and AI engineer. I first published this on 2 July 2026, before the delay became law (dated 8 July, in force 27 July). Dates checked on EUR-Lex on 5 August: check it before you rely on them. The October rewrite is mainly for clarity.
Which date applies to which rule?
IN SHORT Articles 9 to 15 now start in 2027 or 2028, but Article 72 started on 2 August 2026.
IN PLAIN WORDS
The EU AI Act The European law on AI systems, Regulation (EU) 2024/1689. It entered into force on 1 August 2024 and applies in stages. See the glossary has one date when it became law and several dates when its rules start to bind. “In force” means it is law. “Applies” means a given rule now binds.
STEP BY STEP
Step 1. In force. 1 August 2024. The Act becomes law and starts the clock.
- In force 1 August 2024. The Act becomes law and starts the clock.
- The Act applies 2 August 2026. Article 72 monitoring runs from here, as does Article 50 transparency, which can cover systems that are not high-risk: disclosing certain AI uses, flagging AI-generated or manipulated content.
- High-risk, Annex III 2 December 2027, 16 months after the general date. Chapter III, Sections 1 to 3, applies to systems high-risk through a use listed in Annex III (Article 6(2), with exceptions), such as hiring or credit scoring.
- High-risk, Annex I 2 August 2028, 24 months after the general date. The same rules for products, or their safety parts, that Annex I law sends for a third-party check (Article 6(1)), such as toys. For some Annex I laws, only part of these rules applies.
Article 72 covers Post-market monitoring Watching a system once it is in use, and acting on what you find. Article 72 covers it. See the glossary. It sits in Chapter IX, which was never carved out, so it runs on the general date: zero extra months. It binds the provider of a high-risk system.
But the rules that decide what counts as high-risk sit in the chapter that moved. So what Article 72 asks of you before 2027 is a question for your lawyer, not for me.
ITEM GET
Not a free pass
The monitoring, the tests and the docs should grow together.
Regulation (EU) 2026/1744, the Digital Omnibus on AI, is dated 8 July 2026, published on 24 July and in force since 27 July. Point (c) sets the high-risk dates, Article 6(5) excepted. Two more points changed.
Point (a): some provisions newly added to Article 5 apply only from 2 December 2026. Point (d): Articles 102 to 110 apply from 27 July 2026.
A new Article 111(4) covers systems that make synthetic sound, images, video or text, and were placed on the market before 2 August 2026. Their providers have until 2 December 2026 to meet Article 50(2). That rule asks for output marked so software can detect it as AI-made or manipulated.
Sources: consolidated AI Act on EUR-Lex (27 July 2026), Regulation (EU) 2026/1744 in the Official Journal. Check EUR-Lex before you rely on any of it: one unofficial tracker still showed the old text on 5 August 2026.
STAGE CLEAR
What you now know
- You can tell “in force” from “applies”, and place each rule on its clock.
- You know the dates don’t settle whether a system is high-risk: that’s a lawyer’s call.
Check yourself
Tap or click a question to see the answer.
A vendor tells you: “The AI Act was delayed to 2027, so monitoring can wait.” What’s wrong with that?
The Act has applied since 2 August 2026, and Article 72 kept that date. Whether it binds a given system is a question for a lawyer, not a vendor.
Former CTO of Integra. Named author (1 of 5) of ERC-3643, first author of ERC-6960, co-author of ERC-7410, and co-author of ERC-8203, which is still a draft. Building production AI and regulated Web3 systems.
Have a project to ship or a role to fill?
Discuss your project or roleFollow the engineering notes
Get new articles on production AI and smart contracts by email.