In short

Rui Codex builds the technical measures for the transparency obligations in Article 50 of the EU AI Act: accessible notices that tell people they are talking to an AI, structured conversation logs in the EU region you choose, and C2PA and IPTC marking of AI-generated images. Each package is built into your existing software and ends with an evidence pack for your adviser. Not legal advice.

We work on site in Belgium and the Benelux and deliver projects remotely across Europe, including Germany, Austria, Switzerland and France, in English, Dutch, French, German or Turkish, with the GDPR applied to all work.

Does This Sound Like Your Business?

Article 50 of the EU AI Act sets transparency obligations for certain AI systems, which have applied since 2 August 2026. The gap is usually in the software, not in the policy.

Your chatbot never says it is an AI

Visitors chat with your assistant on the website, in WhatsApp or on the phone, and nothing tells them they are talking to software rather than to a person, or the notice exists on one channel and not on the others.

No record of what the assistant said or did

When a customer disputes an answer, there is no structured log of the conversation, the model version, the sources it used or the actions it took, only scattered provider logs that expire.

AI images leave without any marking

Images and media generated in your product or by your marketing team are exported and published with no machine-readable mark, so nobody can later check how a file was made.

Your adviser asks for evidence you cannot produce

Your legal or compliance adviser asks which notice text was live on which date, where logs are kept and for how long, and the answers sit in people's memories rather than in a document.

Your Adviser Decides. We Build It.

The AI Act describes obligations; your legal or compliance adviser decides what they mean for your systems: whether you are a provider or a deployer, which outputs are in scope and which exceptions apply. Rui Codex does the engineering that follows from those decisions, and documents it.

We work in two fixed-scope packages, the AI chatbot transparency and logging kit and AI-generated content marking, built into your existing software rather than bolted on beside it. Each ends with an evidence pack your adviser can review.

Two fixed-scope packages
  • Chatbot transparency & logging kit

    Accessible AI notice in English, Dutch, French and German, structured conversation and action logs in the EU region you choose, evidence pack.

  • AI-generated content marking

    C2PA signing and IPTC metadata in your output pipeline, a visible label, a verification page and tests that the marks survive export.

Each package is scoped for one existing system and agreed in writing before work starts; extra channels, video or audio are added to the scope separately.

Key Capabilities

The technical measures, built, tested and documented in your own software.

  • Accessible AI Disclosure Notice

    A first-message notice and a persistent label that tell people they are interacting with an AI system, readable by screen readers and keyboard users, with a spoken version for voice assistants, in English, Dutch, French and German.

  • Structured Conversation and Action Logs

    Every conversation and every tool call logged with time, channel, model version and sources, in a store you control in the EU region of the hosting provider you choose, with a retention period your adviser sets and CSV or JSON export.

  • C2PA Content Credentials

    Each AI-generated image or media file receives a signed C2PA manifest in your output pipeline, recording that it was produced by an AI system, using open-source C2PA tooling and a signing key held in your organisation's name.

  • IPTC Metadata and Visible Labels

    The IPTC digital source type for AI-generated media written into each file, and a visible label in your CMS or app using the EU icon set for AI-generated content, with a reviewer field where editorial review applies.

  • Verification Page and Survival Tests

    A page or endpoint where anyone can check a file's marks, and tests that send marked files through your own export, resizing, CMS and CDN chain and report which marks survive.

  • Evidence Pack for Your Adviser

    Every notice text version with the date it went live, configuration history, screenshots per channel, test reports and a hand-over note listing what was implemented, so your adviser reviews facts rather than assurances.

How We Work

Five steps from your adviser's requirements to a documented hand-over.

  1. 1

    Intake

    We start from your adviser's requirements: which channels and outputs are in scope, the notice wording, retention periods and any exceptions they have decided.

  2. 2

    Map

    We list every chat entry point and every generation path and output format, and mark each gap against those requirements.

  3. 3

    Build

    We implement the notice, logging and marking inside your existing software, on your infrastructure or in the cloud region you choose.

  4. 4

    Test

    Accessibility checks of the notice with a screen reader and keyboard, log completeness checks, and strip tests of the marks through your real export chain.

  5. 5

    Hand over

    You receive the code, the configuration and the evidence pack, and your adviser receives a note of what was implemented to review and sign off.

Real-World Use Cases

Where these measures typically go.

  • Website Chatbot

    A customer-service or sales assistant on your website gets a first-message notice, a persistent AI label and a structured log of every conversation.

  • WhatsApp, Kiosk and Voice Assistants

    The same notice adapted to each channel: a text notice in messaging, an on-screen label at a kiosk and a spoken notice at the start of a phone call.

  • Chatbots in Public Tenders

    Public tender specifications for chatbots can ask that the AI character is recognisable at all times, that agent actions are logged in structured form and that hosting is in the EU. We build those parts.

  • Image Generation in a SaaS or App

    A feature that generates or edits images for your users signs each output with a C2PA manifest and IPTC metadata, and offers a verification page.

  • AI Images and Video in Marketing

    A marketing team publishing AI-generated visuals gets a label field in the CMS using the EU icon set, and a reviewer log where editorial review applies.

Technology We Use

Open specifications and tools, chosen per project with you.

C2PA c2patool / c2pa-rs IPTC Photo Metadata XMP WCAG 2.2 WAI-ARIA ASP.NET Core PostgreSQL SQL Server Docker OpenAI / Anthropic

Signing keys, any invisible-watermark service and the hosting region are chosen by you and stay in your organisation's name. The rest of our stack is on our technology and methodology page.

Frequently Asked Questions

Common questions about the technical side of the AI Act.

No. Not legal advice: we implement the technical measures your legal or compliance adviser specifies. Whether Article 50 applies to your system, whether you are a provider or a deployer and which exceptions apply are your adviser's decisions. We build, test and document what they specify.

The transparency obligations of Article 50: telling people they are interacting with an AI system, marking AI-generated content in a machine-readable way and labelling published AI content visibly. We do not classify AI systems, write policies or carry out risk assessments.

Yes, that is the usual case. We add the notice layer and the logging to the existing assistant, channel by channel, without rebuilding it. Building a new chatbot is scoped separately.

In a store you control, in the EU region of the hosting provider you choose, kept for the retention period your adviser sets and exportable as CSV or JSON. The logs stay in your environment.

C2PA attaches a signed manifest to the file that records how it was made, and IPTC metadata records its digital source type, such as trained algorithmic media. The European Commission's Code of Practice on transparency of AI-generated content, published in its final version on 10 June 2026, is a voluntary framework; your adviser decides which marking layers you use.

Not always, and we test rather than assume. Some platforms strip metadata on upload. We send marked files through your own export, resizing, CMS and CDN chain and report which marks survive; where your adviser asks for more, an invisible watermark from a provider you choose can be added.

Yes. We can implement the measures in your code base, or work alongside your own developers or the vendor of your chatbot or generation tool, and hand over the code, configuration and documentation.

Article references: Regulation (EU) 2024/1689 (the AI Act), as amended by Regulation (EU) 2026/1744, on EUR-Lex.

Delivered work

Real projects rather than scenarios - what was built, and how.

On our own website

The assistant on ruicodex.com opens every new conversation by introducing itself as the Rui Codex AI Assistant, in all five languages of the site, and our privacy notice names the AI provider behind it.

Where we work

Rui Codex is a Belgian company and works on-site with clients across Brussels, Antwerp, Ghent, Leuven and Liège, serves clients across the Benelux (Belgium, the Netherlands and Luxembourg), and delivers projects remotely across Europe. See all service areas.

Outside the Benelux: how we work with companies across Europe.

Need the Technical Side of Article 50 Built?

Send us your adviser's requirements, or tell us which chatbot or generation feature you run. We will come back with a scoping proposal.

Not legal advice. No commitment. You keep the code, configuration and evidence pack.