Compliance & governance
EU AI Act Transparency Rules: The 2 August Deadline
Most UK firms read the headlines about the EU AI Act being delayed and quietly closed the file. The delay was real, but it was selective. The transparency obligations in Article 50 kept their original date of 2 August 2026, and they are the part most likely to catch an ordinary business.
By Jason Long · July 2026 · 7 min read
The short version
- The headlines said the EU AI Act was delayed. The transparency rules in Article 50 were not: they apply from 2 August 2026.
- The Omnibus simplification package deferred the heavy high-risk obligations to 2 December 2027 (Annex III) and 2 August 2028 (Annex I). Article 50 kept its original date.
- Four duties bite: tell people when they are talking to an AI, machine-mark AI-generated content, tell people when emotion-recognition or biometric categorisation is used on them, and label deepfakes and AI-written public-interest text.
- Breaching Article 50 carries fines up to €15 million or 3% of worldwide annual turnover, whichever is higher (Article 99(4)(g)).
- It reaches UK firms. The Act applies where AI output is used in the EU, wherever the firm sits.
Was the EU AI Act delayed? Partly, and not the part that affects most firms. The Omnibus simplification package pushed the heavy high-risk obligations back to 2 December 2027 for stand-alone Annex III systems and 2 August 2028 for AI embedded in regulated products, but the transparency obligations in Article 50 still apply from 2 August 2026, with fines up to €15 million or 3% of worldwide annual turnover under Article 99(4)(g).
We have had a run of the same conversation this month. A firm sees "EU AI Act deadlines extended", concludes there is nothing to do until 2027, and misses the fact that the extension covered recruitment scoring and medical devices, not the chatbot on its own website. Those are different rules on different clocks.
What was actually deferred, and what was not
The Omnibus reached provisional political agreement on 6 May 2026, was confirmed by Member States on 13 May 2026, and was endorsed by the European Parliament on 16 June 2026 with the Council's final green light on 29 June 2026. It moves the high-risk timetable and simplifies parts of the regime.
It did not move Article 50. The general-purpose AI model obligations that took effect on 2 August 2025 also stand. The single piece of transitional relief in this area is narrow: the machine-marking duty in Article 50(2) gets until 2 December 2026 for generative systems already on the market.
The distinction worth writing down
High-risk classification is about what the AI decides. Transparency is about whether people know AI is involved at all. A firm can be nowhere near high-risk and still owe four transparency duties from 2 August.
The four duties in Article 50
- Direct interaction disclosure (providers). Systems intended to interact with people must be built so users are informed they are dealing with an AI, at the latest at the point of first interaction. Chatbots, virtual assistants and automated phone handling all sit here. The exemption only covers cases where AI involvement is obvious to a reasonably well-informed person.
- Machine-readable marking of synthetic content (providers). Generated or manipulated audio, image, video and text must be marked in a machine-readable format and detectable as AI-generated. Purely assistive editing functions are out of scope.
- Emotion recognition and biometric categorisation (deployers). You must inform the people exposed to the system. If you are running sentiment analysis on customer calls or video, read this one twice.
- Deepfakes and public-interest text (deployers). AI-generated or manipulated image, audio and video content must be disclosed as such. AI-generated text published to inform the public on matters of public interest must be labelled, unless it has been through human review with someone holding editorial responsibility.
Why a UK firm cannot file this under "EU problem"
The Act reaches providers who place AI systems on the EU market and providers or deployers in third countries where the system's output is used in the EU. That is a wider net than "do we have an EU office". A UK professional-services firm with EU clients, an EU-facing support chatbot, or AI-generated content distributed into the EU can be in scope while being entirely UK-registered.
This is the same extraterritorial logic we set out in our guide to the EU AI Act for UK SMEs. What has changed since is that one slice of it now has a firm date two weeks away.
What to do before 2 August
For most firms this is a short, finishable piece of work, and it is governance rather than engineering.
- List every AI touchpoint that faces a person. Chatbots, automated phone systems, AI email replies, call analytics, AI-generated images and video in marketing. If you do not already keep an AI register, this is the moment to start one.
- Fix the disclosures. A single clear line at the start of a chat session, before the first exchange, covers the most common obligation. Retro-fitting it after a complaint does not.
- Decide your labelling rule for content. Write down which AI-assisted content gets human editorial review and who owns that responsibility, because that is what the text exemption turns on.
- Check the tools, not just your own build. Where a vendor is the provider, the marking obligation sits with them; you still need to know whether they have met it.
- Put it in the policy. Add the disclosure and labelling rules to your one-page AI policy so staff have one place to look.
The most common blocker we find is not the rule, it is visibility. Firms cannot label what they do not know is running, and the AI that gets missed is almost always the AI nobody signed off, the shadow AI a team adopted on a card payment last quarter. Find that first; the disclosures are the easy part.
Our 90-minute AI readiness audit produces exactly that list, every AI touchpoint, who owns it, and where a disclosure or label is now required. If 2 August is a date you would rather not think about, book a call and we will work through it with you.
General information for UK firms, not legal or compliance advice. The Omnibus package's amended dates take effect on formal adoption and Official Journal publication; verify current obligations before you rely on them, and take professional advice where needed. Last updated: 20 July 2026.
Sources
- Regulation (EU) 2024/1689, Article 50 — the four transparency obligations, application date of 2 August 2026, exemptions, and the 2 December 2026 date for already-marketed generative systems — https://artificialintelligenceact.eu/article/50/ and https://artificialintelligenceact.eu/transparency-rules-article-50/
- Regulation (EU) 2024/1689, Article 99(4)(g) and 99(6) — fines up to €15,000,000 or 3% of total worldwide annual turnover, whichever is higher; SME provision — https://artificialintelligenceact.eu/article/99/
- Gibson Dunn, EU AI Act Omnibus Agreement — Postponed High-Risk Deadlines and Other Key Changes — provisional agreement 6 May 2026, Member State confirmation 13 May 2026, Annex III deferral to 2 December 2027, Annex I to 2 August 2028, Article 50 unchanged — https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes/
- Greenberg Traurig, Deepfakes, Chatbots, AI-Generated Text: European Commission Details Transparency Obligations Under the AI Act — Commission draft Article 50 guidelines published 8 May 2026, consultation closed 3 June 2026, Code of Practice on marking and labelling — https://www.gtlaw.com/en/insights/2026/6/deepfakes-chatbots-ai-generated-text-european-commission-details-transparency-obligations-under-the-ai-act
- European Parliament endorsement 16 June 2026 and Council final approval 29 June 2026 of the AI Act simplification package — https://www.lw.com/en/insights/ai-act-update-eu-resolves-to-change-rules-and-extend-deadlines
Frequently asked
Was the EU AI Act delayed?
Partly. The Omnibus simplification package deferred the high-risk obligations: stand-alone Annex III systems move to 2 December 2027 and AI embedded in regulated products under Annex I to 2 August 2028. The Article 50 transparency obligations were not deferred and apply from 2 August 2026. The one narrow exception under discussion is the machine-marking duty in Article 50(2), where existing systems have until 2 December 2026.
Do the EU AI Act transparency rules apply to UK businesses?
Often, yes. The Act has extraterritorial reach: it covers providers placing AI systems on the EU market and providers or deployers outside the EU where the system output is used in the EU. A UK firm running a chatbot for EU customers, or publishing AI-generated marketing into the EU, can be in scope regardless of where it is based.
Does my website chatbot need a disclosure?
If it interacts with people in the EU, yes. Article 50(1) requires that people are informed they are interacting with an AI, at the latest at the point of first interaction. The exemption is narrow: it only applies where the AI involvement would be obvious to a reasonably well-informed person.
What is the fine for breaching Article 50?
Under Article 99(4)(g), up to €15,000,000 or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs and start-ups, Article 99(6) applies the lower of the amount or the percentage.
Do we have to label AI-written blog posts and marketing copy?
Only in a specific case. Article 50(4) covers AI-generated or manipulated text published to inform the public on matters of public interest. Text that has undergone human review or editorial control, where a person or organisation holds editorial responsibility, is exempt. Ordinary product marketing is not the target; unreviewed AI commentary on public-interest matters is.
Related insights
Want this sorted, properly?
Our 90-minute audit leaves you with a one-page action list: three things AI should be doing, what it will cost and what it will save. Keep the report either way.