What applies from 2 August 2026
The EU’s AI regulation (Regulation (EU) 2024/1689, the “AI Act”) is coming into effect in stages. The bans on certain AI practices and the AI literacy duty have applied since February 2025, and the rules for general-purpose AI models since August 2025. On 2 August 2026 the transparency obligations in Article 50 follow. They are not just for developers of large models: they also matter to any business that uses AI to deal with customers, such as a chatbot on your website or an AI phone assistant.
The obligations at a glance:
- Telling people they are talking to AI (Art. 50(1)): anyone interacting with an AI system must be informed, unless it is already obvious to a reasonably well-informed, observant and circumspect person. The duty sits with the provider of the system, but if you embed a chatbot on your site, you should make sure a clear notice is shown.
- Machine-readable marking (Art. 50(2)): providers of systems that generate text, images, audio or video must mark the output in a machine-readable way as AI-generated (for example watermarks or metadata).
- Emotion recognition and biometric categorisation (Art. 50(3)): the people concerned must be informed.
- Deepfakes (Art. 50(4)): if you generate or manipulate image, audio or video content that looks real, you must disclose that it is artificial.
- AI text published to inform the public (Art. 50(4)): if AI-generated text is published to inform the public on matters of public interest, this must be disclosed, unless the text has undergone human review or editorial control and someone holds editorial responsibility.
The information must be given clearly, at the latest at the first interaction or exposure, and must be accessible (Art. 50(5)). Austria’s regulator RTR has a helpful summary of the transparency obligations. Breaches of Article 50 can lead to fines of up to EUR 15 million or 3% of worldwide annual turnover; for SMEs, whichever is lower applies (Art. 99).
As practical guidance, the European Commission published a voluntary Code of Practice on marking and labelling AI-generated content on 10 June 2026. Among other things, it explains how deployers can label deepfakes and AI text.
What about the high-risk rules? The “Digital Omnibus”
The obligations for high-risk AI under Annex III (such as AI used in recruitment or creditworthiness checks) were originally due to apply on 2 August 2026 as well. The “Digital Omnibus on AI” moves these dates. The Council gave its final approval on 29 June 2026 and the act was signed on 8 July 2026. At the time of writing (20 July 2026) it has not yet been published in the Official Journal, and it only enters into force after that. Under the adopted text:
| Area | original date | new date under the Omnibus |
|---|---|---|
| High-risk AI under Annex III | 2 August 2026 | 2 December 2027 |
| High-risk AI in products (Annex I) | 2 August 2027 | 2 August 2028 |
| Marking under Art. 50(2) for systems on the market before 2 August 2026 | 2 August 2026 | 2 December 2026 |
Note that the transparency obligations themselves are not postponed. The chatbot notice, deepfake labelling and disclosure of AI text apply from 2 August 2026. Once the Omnibus is published, check whether any details have changed.
What this means for law firms, tax advisers and businesses
Most organisations are deployers under the AI Act rather than providers: they use ready-made AI tools. Article 50 matters most where AI is visible to the outside world:
- Website chatbot or booking assistant: a line such as “You are chatting with an AI assistant” in the chat window itself, not buried in the terms and conditions.
- AI phone assistant: an announcement at the start of the call that an automated system is answering. If you have your own phone system, this can usually be done in the greeting.
- Newsletters, articles, client briefings: whether a text concerns a “matter of public interest” is not always clear-cut, for example a piece on changes in the law. If AI drafts are reviewed by a qualified person who signs them off and takes responsibility, the exception will usually apply. Document that approval process.
- Images and video: realistic AI images of people or places used in marketing should be labelled as AI-generated.
Law firms and tax advisers have a second issue to consider: client data is subject to professional confidentiality and the GDPR. Before pleadings, tax assessments or accounting data go into an AI tool, you need to know where that data is processed. One option is a filter such as houdini, which replaces personal data with placeholders on your own premises before a prompt reaches the AI model.
Whether and in what role your organisation is affected is something to confirm with your legal adviser or professional body if in doubt. This article is not legal advice.
Checklist before 2 August 2026
- Inventory: which AI systems have contact with customers, clients or the public (chat, phone, email replies, website content)?
- Clarify your role: are you a provider (own development, own name) or a deployer (off-the-shelf product)?
- Add notices: an AI notice in the chat window and the phone greeting, visible and easy to understand.
- Ask your vendors: does the tool mark its output in a machine-readable way? What does the contract say about data processing and storage location?
- Internal AI policy: which tools are allowed, what data may go in, who reviews and signs off, how content is labelled.
- Training: brief your staff on labelling and data protection.
- Plan high-risk projects: if you use AI in recruitment or similar areas, plan for the new 2027/2028 dates.
How IT United can help
As part of our AI consulting, we work with you to map the AI tools in use, check technical settings such as notices, logging and data flows, and help you put together an internal policy that works in practice. The legal assessment stays with your legal adviser; we make sure the technology matches it.
Would you like to know where AI is already in use in your organisation and what needs doing before August? Book a free initial consultation and we will be happy to talk it through.
