EU AI Act deadline: what changes on August 2, 2026
New EU rules require businesses to tell customers when they're talking to AI, and to label AI-generated content. Here's what's required, who it affects, and how to check your own business in under 20 minutes.
Who this is for: Business owners, operations leads, and managers at Irish or UK companies that use AI tools, whether that's a chatbot, an AI writing tool, or AI-powered customer service, whether you built it yourself or bought it off the shelf.
On July 20, 2026, the European Commission published its final guidance on a set of AI transparency rules that become law on August 2, 2026. These rules affect the AI tools most businesses already use: customer chatbots, AI writing tools, AI-generated images or video.
If you've seen headlines saying the EU AI Act was delayed, that's only partly true, and the part that wasn't delayed is the part that applies to most everyday businesses. If you only read the headlines, you could easily assume none of this hits you this year. It does.
Quick summary
- August 2, 2026 is when new AI disclosure rules become enforceable across the EU.
- These rules mainly affect chatbots, AI-generated content, and AI tools that read emotions or biometric data.
- A separate EU reform (the "Digital Omnibus") delayed different, stricter rules for AI used in hiring, credit scoring, and similar high-stakes decisions to December 2027 and 2028. That delay does not cover the rules discussed here.
- Breaking these rules can cost a company up to €15 million or 3% of global annual revenue, whichever is higher.
- UK businesses aren't automatically exempt if they serve EU customers.
Why there's so much confusion right now
Earlier this year, the EU passed a reform package called the Digital Omnibus, which pushed back the deadline for the AI Act's toughest rules, the ones covering "high-risk" AI used in hiring, credit decisions, and similar situations. That delay produced a wave of "AI Act pushed back" headlines.
The problem: most businesses don't run that kind of high-risk AI. They run chatbots, AI writing assistants, and AI image tools, and the rules covering those tools were not delayed. They still take effect on August 2, 2026. We covered the high-risk categories and the full Omnibus background in our earlier post on the EU AI Act; this one focuses on what actually lands in August and a 20-minute triage you can run yourself.
What's actually happening on August 2, 2026
Two separate things kick in on this date.
First, AI transparency rules for everyday AI tools. If your business uses an AI system that talks to customers, generates content, reads emotional cues, or creates realistic AI images or video, you now have specific disclosure duties. These are covered in detail below.
Second, fines for the companies behind AI models. The companies that build the underlying AI models your tools run on have had documentation and copyright rules since August 2025. Starting August 2, 2026, EU regulators gain the power to fine them for breaking those rules.
What got delayed, and what didn't
The Digital Omnibus reform made three changes, and none of them touch the rules above:
- AI used for high-stakes decisions about people, such as hiring, credit scoring, or insurance eligibility, now has until December 2, 2027 to comply, pushed back from August 2026.
- AI built into regulated physical products, like medical devices or industrial machinery, now has until August 2, 2028.
- AI content tools already on the market before August 2, 2026 get a grace period until December 2, 2026 to add machine-readable labels to their output. Anything launched from August 2, 2026 onward must have this in place from day one.
Everything else, the core disclosure rules for chatbots, AI content, and biometric tools, is unchanged and still takes effect on schedule.
Step one: work out if you're a "provider" or a "deployer"
Before you can figure out what applies to you, you need to know which role you play. The EU AI Act splits responsibility into two categories.
A provider built the AI tool, or had someone build it for them, and it runs under their company's name or brand. If an agency built your chatbot and it appears on your website as yours, you're the provider.
A deployer uses someone else's AI tool as-is, like a subscription chatbot or AI writing platform used without customization beyond basic setup.
This matters because certain duties fall on providers, others on deployers, and some apply to both. Sort your AI tools into these two buckets first, since it decides everything that follows.
The 4 disclosure rules you need to know
1. Tell customers they're talking to AI
If a chatbot, voice assistant, or other AI tool talks directly to your customers, they need to know it's AI, clearly, and from the first message. A line buried in your terms of service doesn't count.
In practice that means a visible label like "AI assistant" in the chat window, plus an opening message that says so directly. It costs nothing to add and removes any ambiguity.
If you bought a chatbot from a vendor rather than building it, this responsibility sits with them, but you should still confirm the disclosure is switched on by default and that your branding customizations haven't hidden it.
2. Label AI-generated content
If your business generates AI images, video, audio, or text and that content goes out into the world, it generally needs an embedded, machine-readable marker showing it was AI-made. This applies whether you built the tool yourself or you're using someone else's AI model inside your own product.
There's a narrow exception for tools that only make small assistive edits, like spell-check, but a proper content generator won't qualify.
Here's the practical catch: simple embedded labels can get stripped out. Resizing an image, converting a video file, or copying text out of a document can wipe the marker without anyone noticing. If you rely on a vendor's AI tool, ask them two questions: what labeling method do they use, and does it survive the way you actually use the output?
3. Disclose emotion-detection and biometric tools
If you use any tool that reads people's emotions or physical traits, a common example is AI that analyzes tone of voice on customer service calls, you have to tell the people being analyzed, and you have to handle that data under standard EU data privacy law (GDPR).
One related rule worth knowing: using AI to read your own employees' emotions in the workplace has been banned outright since February 2025, with only narrow medical or safety exceptions. The disclosure rule above covers uses that remain legal, like customer call analysis. It doesn't cover workplace monitoring, which is banned separately.
In practice, if you run call analysis software that detects customer sentiment or tone, tell customers it's in use. That's a line in the call flow and an update to your privacy policy. A silent rollout is exactly what this rule is meant to stop.
4. Label deepfakes and AI-written public content
Two situations trigger this rule. You publish an AI-generated or edited image, audio clip, or video that looks like a real person, place, or event and could be mistaken for genuine footage, sometimes called a deepfake. Or you publish AI-written text intended to inform the public on a matter of public interest, think news-style or public-information content.
Two exceptions matter here. Clearly artistic or satirical content only needs a light-touch disclosure that doesn't undercut the joke. And AI-drafted text is exempt from labeling if a person has actually reviewed it and taken editorial responsibility for it before publishing.
For most businesses, this rule mainly affects marketing and blog content drafted with AI. The simplest path to compliance is having a named person review AI-drafted content before it's published, and documenting that step. If your content goes out unreviewed, you need to either add a review step or label the content as AI-generated.
What happens if you don't comply
Breaking any of the four rules above can result in fines of up to €15 million or 3% of your company's worldwide annual revenue, whichever is higher. Smaller businesses get a lighter version of this cap, but the exposure is still real, and it applies to providers and deployers alike.
Enforcement runs through national regulators rather than Brussels directly. In Ireland, 15 industry-specific regulators now oversee this, coordinated by a new national AI Office. A dedicated Irish AI law is also close to finalized in parliament. Even where national enforcement systems are still being set up, the underlying rules are binding from August 2, 2026. A slow national process doesn't change the EU deadline.
Does this apply to UK businesses?
Yes, more often than most UK business owners expect. The rules apply to any company, anywhere, whose AI tool is used by people in the EU. A UK company with a chatbot that serves EU customers, or whose AI-generated content reaches EU audiences, falls under these rules. The UK doesn't currently have an equivalent law of its own, so for any EU-facing part of your business, these EU rules are the ones that count.
Your 5-step compliance checklist (20 minutes)
- List every AI touchpoint in your business: chatbots, voice bots, AI content or image tools, call-analysis software, AI-assisted publishing.
- Mark each one as provider or deployer. Built under your brand means provider. Bought and used as-is means deployer.
- Test your own chatbot like a first-time visitor. If it's not obvious within the first screen that you're talking to AI, add a label and an opening line.
- Check your AI-generated content. Ask your vendor, or check yourself, what labeling method is used and whether it survives your normal editing process.
- Confirm your content review process. If AI drafts public-facing text, make sure a named person reviews it before publishing. If not, add that step.
Most businesses that run through this checklist end up with a short, manageable list of fixes: a label here, a vendor email there. The businesses with real work ahead are usually those generating AI content with no labeling plan, or those that have quietly added emotion-detection tools without telling anyone.
Need help closing the gap?
We build AI systems for clients across Ireland and the UK, and we track this regulation closely because our own products fall under it too. If this checklist surfaces something you can't resolve before August 2, 2026, book a call or explore our AI consulting practice. Bring your list from step one; the first conversation usually resolves most of it.
Frequently asked questions
What is Article 50 of the EU AI Act?
Article 50 is the part of the EU AI Act that requires businesses to disclose when people are interacting with AI, when content is AI-generated, and when AI tools analyze emotions or biometric traits. These rules become enforceable on August 2, 2026.
Was the EU AI Act delayed?
Only partly. A separate reform, the Digital Omnibus, delayed the rules for high-risk AI used in hiring, credit scoring, and similar decisions to December 2027 and 2028. The disclosure rules for chatbots, AI content, and biometric tools were not delayed and still take effect August 2, 2026.
Do small businesses have to comply with the EU AI Act?
Yes. The disclosure rules apply to any business using qualifying AI tools, regardless of size, though fine caps are adjusted lower for small and medium-sized enterprises.
Does the EU AI Act apply to UK companies?
Yes, if a UK company's AI tools are used by people in the EU, for example a chatbot serving EU customers or AI content viewed by EU audiences.
What's the penalty for not complying with the AI Act's disclosure rules?
Fines of up to €15 million or 3% of worldwide annual revenue, whichever is higher, with a reduced cap for smaller businesses.