The 5-Second Test: Is Your Chatbot in Breach of the EU AI Act Today?

The EU AI Act deadline nobody's chatbot met — and it's not the scary high-risk one.

Article 50(1) has applied since 2 August 2026: if your AI interacts directly with people, those people must know they're talking to a machine. Most compliance programs are busy with the 2027 high-risk deadlines while their own website chatbot fails a live rule right now.

┌─────────────────────────────────────┐
│ THE 5-SECOND TEST                   │
├─────────────────────────────────────┤
│ Open your chat widget in incognito. │
│ First-time visitor. Five seconds.   │
│                                     │
│ Do you KNOW it's a machine?         │
│   No  → you're in scope today       │
│   Yes → document why, then breathe  │
└─────────────────────────────────────┘

What doesn't count as disclosure:

✗ "I'm an AI" buried in the terms page ✗ Gray-on-gray microtext in the widget footer ✗ A human name ("Anna from Support") plus typing-indicator theatrics

What does:

✓ A header line or first message: "You're chatting with our AI assistant" ✓ Visible at first interaction, on every re-entry ✓ Accessible — perceivable, not decorative

The classic enforcement file is a customer complaint with a transcript attached, claiming they were "tricked by a fake human." Exposure: up to €15M or 3% of turnover under Article 99 — and it starts at consumer complaints, not regulator visits.

Fix cost: one header line and one opening message. There's no legitimate reason a post-August chatbot ships without it.

Full guide with the 60-minute audit checklist and the December 2026 content-marking nuance: Article 50 and Your Website Chatbot