Quick answer: Since 2 August 2026, the EU AI Act’s transparency rules apply. Chatbots must tell people they’re talking to a machine, and deepfakes must be clearly labelled. Rules for “high-risk” AI, such as AI used in hiring or schooling, were pushed back to December 2027 and August 2028, and a new ban on “nudify” apps starts on 2 December 2026.
The EU rewrote part of its AI timetable this summer. A package known as the Digital Omnibus on AI was published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026 as Regulation (EU) 2026/1744, according to White & Case. It delayed some obligations but left the consumer-facing transparency rules on schedule.
If you live in the EU, you’ll see the effects first. If you’re in the US, UK, Canada or Australia, the law doesn’t apply to you directly, but it covers AI services offered in the EU, so global products may change how they label content. This is a wider look at the AI news and guides we cover, so treat the labels below as something to watch for, not a promise.
What changed on 2 August 2026?
The European Commission says the AI Act’s transparency rules apply from August 2026. In its own words, people must be “made aware that they are interacting with a machine so they can take an informed decision.” You can read the summary on the Commission’s AI Act page.
Three duties matter most to everyday users:
- Chatbot disclosure: an AI assistant that talks to you must make clear it isn’t a person.
- Machine-readable marking: companies that provide generative AI tools must mark audio, image, video and text outputs so software can detect they’re AI-made.
- Visible labels on deepfakes: people who publish deepfakes, and AI text on matters of public interest, must label them clearly.
What will AI labels look like?
The Commission published a voluntary Code of Practice on marking and labelling AI-generated content in mid-June 2026, as reported by the Society for Computers and Law. It isn’t binding, but it shows how companies can meet the binding rules.
According to Emerging Europe, the code points to digitally signed metadata and imperceptible watermarking for the machine-readable part. For the visible part, it describes a common EU icon with two variants: “fully AI-generated” and “partially AI-modified.”
There are exceptions. Artistic, satirical or fictional deepfakes face lighter disclosure duties. AI-written text doesn’t need a label if a person has reviewed it and takes editorial responsibility for it.
Who has to label what, in plain terms?
Think of it as two jobs. The companies that build AI tools handle the invisible marking, so a detector can tell an image or clip came from a model. The people or organizations that publish the content handle the visible label, when the content is a deepfake or public-interest text.
That split matters for ordinary users. A friend posting an AI-edited holiday photo isn’t the target. A campaign posting a fake video of a politician is. The Commission’s guidance also says certain AI-generated content must be “clearly and visibly labelled,” which is why the code of practice spends so much time on icons.
What got delayed?
The “high-risk” category covers AI that affects big decisions in people’s lives, such as hiring, education, credit and law enforcement. Those rules were due on 2 August 2026. They now apply as follows, per Gibson Dunn:
| Type of system | Old date | New date |
|---|---|---|
| Stand-alone high-risk systems (hiring, education, biometrics and similar) | 2 August 2026 | 2 December 2027 |
| High-risk AI built into regulated products (medical devices, machinery) | 2 August 2027 | 2 August 2028 |
| Watermarking for tools already on the market before 2 August 2026 | 2 August 2026 | 2 December 2026 |
So for now, an employer using AI to screen CVs in the EU has more time before the strictest rules bite. Existing image and video generators have until 2 December 2026 to start marking their output.
What is the new ban on nudify apps?
The omnibus added a prohibition on AI systems designed to create non-consensual sexual or intimate imagery, including child sexual abuse material. Law-firm summaries from Winston Taylor and Gibson Dunn agree that it covers so-called nudification apps and that it applies from 2 December 2026, with limited exclusions. Providers of general image and video generators also have to assess foreseeable misuse risks.
For you, that means a legal basis for pushing app stores and providers to remove these tools in the EU. It doesn’t make fake images disappear overnight, and enforcement will take time. If you’ve been targeted by fake images, our guide on how to spot and report them is a good next step, and our piece on https://insightorys.com/2026/10/07/use-ai-safely-at-work/ covers what not to paste into any chatbot.
What happens if a company ignores the rules?
Emerging Europe reports that fines for breaching the transparency rules can reach 15 million euros or 3 percent of worldwide annual turnover, whichever is higher, with a lower cap for small and medium-sized businesses. Individual users aren’t fined for sharing a deepfake casually; the duty falls on the providers and on those who publish the content.
What should you do now?
- Treat an “AI” label as a hint, not a verdict. Labelling depends on the publisher, and bad actors won’t comply.
- If you publish AI-made images or video that look like real people or events, label them. This applies to creators and small businesses selling in the EU.
- Check whether your favorite AI tool now shows a “you’re talking to an AI” notice, and note which ones include content credentials in exported images.
- If you use AI for work decisions about people, keep notes on how you use it. The strict rules arrive in 2027, but the habit helps now.
Frequently asked questions
Does the EU AI Act apply to me if I live in the US, UK, Canada or Australia?
Not directly. It’s EU law and covers AI services offered to people in the EU. Companies that sell globally may still apply the same labels everywhere, but that’s their choice, and we haven’t seen a confirmed worldwide rollout.
Is the EU AI Act delayed?
Only parts of it. High-risk obligations were delayed to December 2027 and August 2028. The transparency rules for chatbots and deepfakes still started on 2 August 2026, with a grace period to 2 December 2026 for watermarking on existing tools.
Will every AI image carry a visible label?
No. The visible label duty targets deepfakes and certain public-interest text. Most AI output gets invisible, machine-readable marking instead, which you can’t see without a detection tool.
Is the Code of Practice mandatory?
No, it’s voluntary. It’s a way for companies to show they’re meeting the binding AI Act duties.
Sources
- European Commission: Regulatory framework on AI
- White & Case: EU AI Omnibus enters into force, amending the AI Act
- Gibson Dunn: EU AI Act Omnibus Agreement, postponed high-risk deadlines and other key changes
- Winston Taylor: AI Act rules on high-risk AI delayed as AI Digital Omnibus agreed
- Emerging Europe: Europe’s AI labelling rules arrive
- Society for Computers and Law: European Commission publishes Code of Practice on marking and labelling AI-generated content