← Blog
#AI#LLM#Compliance#Security#Architecture

EU AI Act Article 50 Is Live: What Chatbot and Agent Builders Need to Implement

webhani·

Article 50 of the EU AI Act — the transparency obligations section — took effect on August 2, 2026. The European Commission adopted its guidelines on July 20. Unlike the "high-risk system" classification that dominated earlier AI Act coverage, Article 50 applies broadly: chatbots, voice assistants, AI agents, and AI-generated content all fall under it. Penalties run up to €15 million or 3% of global annual turnover, whichever is higher.

webhani builds AI-powered web applications, and this regulation applies based on whether you're serving EU users — not where your company is based. Here's what the article actually requires from an implementation standpoint.

The Four Scenarios Article 50 Covers

The article splits obligations between providers (who build and place the AI system on the market) and deployers (who use it under their own authority), across four scenarios:

  1. Systems that interact directly with people (chatbots, voice assistants, AI agents) — must disclose that the user is talking to AI, unless this is already obvious
  2. Emotion recognition or biometric categorization systems — must notify the person they're subject to it
  3. AI-generated or manipulated content (including deepfakes) — must be clearly marked as artificial
  4. AI-generated text on matters of public interest — must carry machine-readable markings and be detectable

For most web teams, scenarios 1 and 4 are where the real implementation work is.

Pattern: Disclosure for Chatbots

"Unless already obvious" is vague enough that the safe interpretation is a persistent, explicit label in the UI — not a one-time modal that disappears.

// components/ChatDisclosure.tsx
export function ChatDisclosure() {
  return (
    <div
      role="status"
      aria-live="polite"
      className="text-xs text-muted-foreground border-b px-3 py-2"
    >
      You're chatting with an AI assistant.
    </div>
  );
}

Don't rely solely on a session-start banner. For long-running support chats, keep the disclosure visible even if a user joins the conversation midway — a header label survives scroll and re-renders better than a toast.

Pattern: Marking AI-Generated Content

Article 50(2) requires machine-readable marking for AI-generated text and images. The May 2026 AI Omnibus provisional agreement gives generative systems already on the market a grace period until December 2, 2026 for this requirement — but new builds should implement it from day one.

C2PA (Coalition for Content Provenance and Authenticity) Content Credentials are becoming the de facto standard for cryptographically signed metadata on images and video. Text watermarking standards are still less settled, so a pragmatic starting point is explicit signaling via response headers or structured metadata.

// app/api/generate/route.ts
export async function POST(req: Request) {
  const result = await generateText({ /* ... */ });
 
  return Response.json(
    { content: result.text },
    {
      headers: {
        "X-AI-Generated": "true",
        "X-AI-Generator": "webhani-assistant/1.0",
      },
    }
  );
}

Since C2PA-style provenance standards will likely become expected practice beyond just text, it's worth designing your generation pipeline with an extension point for embedding provenance metadata now, rather than retrofitting it later.

Our Take

It's tempting to treat this as compliance overhead bolted on at the end. In practice, it's cheaper to treat disclosure and provenance marking as part of the component contract from the start. Retrofitting a disclosure label or metadata field after the fact usually means touching component structure and API response shapes across the codebase.

For any new AI-powered project, we recommend baking "disclose the AI interaction" and "mark generated content" into your component standards during initial design — even for products outside the EU. Similar transparency requirements are spreading to other jurisdictions, so designing for it now avoids a second retrofit later.

Takeaways

  • Article 50 has been in effect since August 2, 2026, and applies well beyond "high-risk" systems
  • Chatbots and agents must disclose that users are interacting with AI
  • AI-generated content needs machine-readable marking (existing systems get until December 2, 2026)
  • Build disclosure UI and provenance metadata into your design phase, not as an afterthought

We help teams design AI features with this kind of compliance built in from the start — reach out if you want a second set of eyes on your implementation.

← Back to Blog