What happened. As of August 2, 2026, the transparency obligations under Article 50 of the EU AI Act will be fully applicable. In May 2026, the EU Commission opened a public consultation on the final guidelines and simultaneously published the accompanying Code of Practice for transparent AI systems for comment. This is the most important legal change of the summer — and the most tangible in practice for any company deploying AI to end users.
Context
Article 50 requires four things simultaneously, and they apply to different use cases. First: AI systems that interact directly with natural persons — chatbots, voice assistants, AI avatars — must disclose that they are AI. This obligation falls on providers, not deployers. Second: providers of generative AI systems must technically mark their synthetic content as AI-generated — typically via C2PA metadata or digital watermarks. Third: deployers who publish deepfakes or machine-generated text on matters of public interest must label them as such. Fourth: deployers of emotion recognition or biometric categorization systems must inform affected individuals.
The EU Commission’s May 2026 consultation primarily addresses how these four obligations are to be documented, verified and enforced. The guidelines will clarify what a “clear marking” concretely means, whether it must be machine-readable, in what format watermarks must be provided, and how providers and deployers divide responsibility for compliance.
What this means in practice
For Lumi-Systems.io and comparable providers with their own web offerings, this means specifically: every chatbot on the website must be clearly labeled as AI from August 2026 onwards. Every AI-generated article published by us must be marked as such — we have done this on Lumi AI News since launch with a notice at the end of each imported or AI-curated article, and this practice standard will become mandatory from August. Every generative AI service that we in turn provide to customers as a provider must technically mark its outputs.
Mid-market companies deploying AI should use the summer to conduct three inventories: where does AI interact with end users (chatbot inventory), where does AI generate content for the public (content inventory), where is biometric data or emotion processed (sensor inventory). Each of these three categories has its own marking obligation from August 2 — and each violation is subject to penalties of up to €15 million or 3% of global group annual turnover, whichever is higher.
The consultation runs through summer 2026. Associations, law firms and critical stakeholders can still have influence. Those wishing to submit arguments against unclear definitions or impractical watermarking requirements have the window until publication of the final guidelines.
Original sources
- EU Commission — Consultation on Transparency Guidelines (May 2026)
- EU Commission — FAQ Guidelines & Code of Practice on transparent AI
- EU AI Act — Regulatory Framework
- Bird & Bird — Draft Transparency Code of Practice (Analysis)
- EU AI Act Tracker — Implementation Timeline
— Lumi AI Act Watch · May 25, 2026. Research and first draft AI-assisted, editorial approval by Lumi-Systems.io. Marked in accordance with Article 50 EU AI Act.