Back to blog
AI / GEO·7 min read

EU AI Act Article 50: are you ready for August 2026?

Hard deadline 2 August 2026: every website with a chatbot or AI-generated content must disclose it. Penalties up to €35M or 7% global revenue. Here's the technical checklist.

The EU AI Act entered into force on 1 August 2024 with phased compliance deadlines. Article 50 — covering transparency obligations for AI providers and deployers — becomes binding on 2 August 2026. Penalties: up to €35M or 7% of global annual turnover, whichever is higher.

What Article 50 actually requires

Three concrete obligations for website operators:

  1. Chatbot disclosure: if a natural person interacts with an AI system (chatbot, voice agent, support widget), they must be informed clearly they're talking to a machine. No buried disclaimers.
  2. AI-generated content labeling: synthetic audio, image, video, or text generated or manipulated by AI must be marked in a machine-readable way (e.g., C2PA Content Credentials).
  3. Emotion recognition / biometric categorisation disclosure: if you use these systems on visitors, you must inform them.

How to comply (technical checklist)

  • Add visible disclosure to every chatbot widget: "You are chatting with AI" (label, tooltip, or first message). Don't hide in footer ToS.
  • For AI-generated images/videos, embed C2PA Content Credentials metadata. Most modern image-gen tools (Adobe Firefly, OpenAI DALL-E 3, Microsoft Designer) embed automatically — but verify the metadata survives your CDN.
  • Add an /legal/ai-notice page documenting which AI systems you use, for what purposes, with which models, and where data is processed (especially relevant for EU sovereignty claims).
  • For chatbots: ensure the first turn of any conversation starts with the AI disclosure, not buried at end of welcome message.

What about RAG / AI-assisted content?

The Act is specific: "AI-generated or manipulated". A human-written article that uses an AI tool for spell-check, translation, or summarization doesn't require labeling. But: a synthesis page assembled by RAG from documents, or a blog article generated end-to-end by GPT/Claude, does require disclosure.

Auditope check

Our P3.1 phase (shipped v0.19.0) detects: chatbot widgets via DOM patterns (Intercom, Drift, Tidio, custom React components), AI disclosure presence/absence in chatbot text, and C2PA metadata on visible images. Findings emitted with severity HIGH and direct quote from Article 50 in the recommendation field.

You have ~9 weeks until enforcement. Run a check now at auditope.com.

Want this kind of analysis on your own site?

Run a free audit →