Most of the coverage of the Digital Omnibus concentrated on the sixteen-month delay to the AI Act's high-risk regime. Less noticed is what the Omnibus left in place. Article 50, the transparency article, applied on 2 August 2026 to new systems as originally scheduled; systems already on the market before that date have until 2 December 2026. For professional firms, whose AI systems are rarely high risk but frequently talk to clients, Article 50 is the part of the Act that matters this year.
1. What Article 50 asks
- Interaction. People interacting with an AI system must be informed that they are, unless it is obvious from context to a reasonably well-informed person.
- Generated content. Providers must ensure AI-generated or manipulated text, audio, image and video outputs are marked as such in a machine-readable way.
- Published text. Deployers who use AI to generate text published to inform the public on matters of public interest must disclose it, unless the text has been through human review and editorial responsibility.
The Omnibus adjusted timing and softened the separate AI-literacy duty, but the substance above is unchanged.
2. Where it bites in a firm
The obvious case is a client-facing assistant on a portal or a phone line: that needs a clear statement. The less obvious cases are the ones agents create. An automated request to a client for a missing bank statement. A first-draft status update on a matter. A summary of a document sent to a counterparty. Each is text generated by an AI system that reaches a person. Where a qualified professional has reviewed it and taken responsibility for it, the firm is on solid ground. Where it went out unreviewed, the firm should be able to show how the recipient could tell.
The Act does not ask firms to apologise for using AI. It asks them not to pretend they did not.
3. Doing it well
- Decide the line. Which communications go out with human review and sign-off, and which go out automated. The first group is professional work product; the second needs a disclosure.
- Make the disclosure plain, not defensive. "This update was prepared automatically from your matter file; your adviser reviews the file weekly" reads as service. A paragraph of legal boilerplate reads as a warning.
- Keep the receipt. The workflow should record which path each communication took, so the firm can show the regulator, the insurer or the client which messages were reviewed and by whom.
- Align with professional rules. Bars and institutes already expect candour about how work is produced. The Act adds a floor, not a new philosophy.
4. The opportunity in it
Firms in the DACH region and elsewhere in Europe have found that explaining how AI is used, plainly and early, wins work. Clients in regulated industries have their own obligations and would rather engage a firm that has already thought this through. A short, honest statement of how the firm uses AI, what is reviewed and how data is handled turns a compliance duty into part of the pitch.
Hive Newsroom follows what is changing in AI and professional services. Sources are linked in the text; figures are as published at the time of writing. Comments and corrections: press@get-hive.ai. Back to the .