In one sentence
From 2 August 2026, the transparency obligations under Art. 50 of the EU AI Act take effect – for law firms, this doesn't mean stamping every AI-assisted letter, but labelling wherever clients or courts would otherwise be misled about the origin of an interaction or a piece of content.
When does your firm become a "deployer" under Art. 50?
The AI Act addresses law firms as "deployers" the moment they use an AI system. What matters, though, is the type of use — not every use triggers the labelling requirement:
- Firm chatbot on the website: if a system interacts directly with people, it must be disclosed that it's AI — unless that's obvious anyway.
- AI-generated client information: synthetically generated text, images, or audio distributed as information must be labelled in a machine-readable way and recognisable to the recipient.
- AI-assisted pleadings: the pleading you review, take responsibility for, and sign is your own work product as a lawyer. Using a tool in the drafting process doesn't, by itself, trigger a labelling obligation toward the court — final professional responsibility remains with you.
Exactly where the line runs is still an open interpretive question that will only become sharper through regulatory practice and case law. You'll find the full text in the AI Act on EUR-Lex.
The AI Act meets professional conduct rules
Much of what the AI Act now codifies is already required by attorney professional-conduct rules anyway. The BRAO's obligation to practise the profession diligently and on one's own responsibility means: you check every AI output before it leaves your firm. Courts have repeatedly made clear that invented citations or references from an AI system breach this duty of care and can lead to a formal reprimand (more on this in our assessment of the Berlin hallucination case).
The AI Act also creates an additional, public-law layer: labelling toward third parties. Professional-conduct rules protect the client relationship, the AI Act protects the recognisability of AI. Both layers run in parallel — one doesn't replace the other.
Checklist ahead of the deadline
- Take stock: which AI systems are in use at your firm — chatbot, text generators, research tools? And for exactly what purpose?
- Labelling concept: determine which outputs go out externally and need a notice, and where your professional responsibility as the lawyer makes labelling unnecessary.
- Demand provider transparency: ask your tools for information on the models used, the hosting region, and subprocessors. Without this information, you can't properly meet your own labelling obligation.
According to current surveys, 63.6% of German law firms actively use AI — two years ago it was under 30%, and more than 60% of small firms are on board too. So the question is rarely "whether" but "how transparently".
How our Anwaltsportal reflects this
We've firmly anchored transparency at several points:
- Model transparency in the picker: you can see at any time which model generated a given turn.
- EU pinning even for the premium model: even the Fable add-on runs with EU pinning and only after an explicit firm-level opt-in.
- Public subprocessors list: our subprocessors list and our AI transparency page are openly accessible — exactly the information you need for your own labelling concept. Your case data doesn't flow into model training.
You'll find more on the portal at /fuer-anwaelte.
Not a ban — a transparency requirement
The AI Act doesn't sanction the use of AI as such. Violations of the transparency obligations can be punished with fines of up to €35 million or 7% of worldwide annual turnover — so what's punished is a lack of transparency, not the use of modern tools. For high-risk applications under Annex III, a later deadline also applies: these obligations were postponed to 2 December 2027 via the "Digital Omnibus". Anyone who documents cleanly today where and how AI is used is prepared for both stages.
This article provides general orientation and doesn't replace legal advice in an individual case.
