In one sentence
From 2 August 2026, chatbots must disclose that they are AI, and AI-generated content as well as deepfakes must be labelled — the stricter high-risk obligations, by contrast, do not arrive until 2027.
The deadline: 2 August 2026
With the EU AI Act 2026, the transparency obligations under Article 50 of the AI Regulation become applicable from 2 August. The core idea is easy to remember: if you are communicating with an AI, you should know it. In practice, this means three things:
- Chatbots and AI assistants must reveal that they are AI, unless this is already obvious.
- AI-generated text, images, audio and video content must be marked as such in a machine-readable way.
- Deepfakes — deceptively realistic, artificially generated content — need a clear notice of their artificial origin.
The scope for fines is considerable: breaches can be penalised with up to €35 million or 7% of worldwide annual turnover — whichever is higher. This labelling requirement is therefore anything but a non-binding suggestion.
What is NOT arriving in August
This is where most of the current confusion arises. Many expect that from August "the whole AI Act" will apply. That is not the case. The demanding obligations for high-risk AI systems under Annex III — for example in recruitment, lending or the justice system — have been postponed to 2 December 2027 as part of the so-called Digital Omnibus.
The clean distinction matters: only the high-risk requirements have been postponed. The transparency and labelling obligations under Article 50 take effect independently from 2 August 2026. So anyone operating a chatbot or publishing AI content cannot rely on 2027.
What this means for you as a consumer
For you, one thing above all changes for the better: in future you will be told more reliably when you are talking to a machine and when a piece of content was generated by an AI. This makes healthy scepticism easier — especially on legal or financial topics, where an AI answer is good orientation but not verified information.
What you as an SME should do now
If you run a website with a chatbot or publish AI-generated text or images, a quick check before August is worthwhile:
- Does your chatbot clearly identify itself as AI — visibly, not just in the small print?
- Is AI-generated content on your site recognisable as such?
- Is there an internal person responsible who knows which AI tools are running in-house?
- Is it documented which providers and models you use?
These four points are not a complete compliance programme, but an honest starting point. We explain the precise terms — from "AI system" to "deepfake" — on an ongoing basis in the glossary.
How SmartLegalPro already works today
We did not wait for the labelling rules; we built them in from the start. Across all our surfaces, it is visible where an AI is working and where a human decides. The models used are transparently selectable, our data processing runs on EU hosting, and we do not train on client data. If you want the full detail, you will find it on our AI transparency page and in our openly disclosed subprocessor list. Tools such as PII masking and PDF redaction follow the same principle: traceable rather than a black box.
And the limit
Helpful as an overview may be, it does not replace an assessment of your individual case. Our AI provides a 60-second initial assessment with no account and takes the groundwork off your hands; whether your particular website, chatbot or product is AI Act-compliant, however, belongs in the hands of a lawyer. That is exactly what our one-click handover to pre-qualified law firms is for.
