EU AI Act for Dottie operators
A short deployer briefing: what Article 50 transparency means for voice agents, what staff must know (Article 4 literacy), and what Dottie already enforces vs what remains your responsibility.
- EU AI Act
- compliance
- transparency
- voice AI
Auf einen Blick
Skim the essentials, then dig into the steps.
- Tell people it’s AICallers must know they are speaking with an AI assistant at the first interaction.
- Train your teamAnyone who configures or monitors the agent should understand limits and handover.
- Keep disclosuresDo not strip AI wording from greetings or market the agent as a human.
- Stay in scopeBusiness voice help is fine; prohibited and high-risk uses need a different legal path.
The EU AI Act is risk-based. For most small businesses using Dottie as a phone or browser voice assistant, the obligations that bite day-to-day are transparency (Article 50) and AI literacy (Article 4) — plus GDPR for personal data. High-risk rules (Annex III) and prohibited practices (Article 5) are out of Dottie product scope by design and banned in the Acceptable Use Policy.
Section 1
Who is the provider and who is the deployer?
- Dottie is the provider of the voice agent system: we design defaults so callers are told they are speaking with AI.
- Your organisation is typically the deployer: you decide when the agent runs, what knowledge it uses, and how staff supervise it.
- Callers’ personal data is usually under your controllership; Dottie processes it under the DPA.
Section 2
Article 50 — transparency in practice
When an AI system interacts directly with people, they must be informed they are interacting with AI, in a clear way, at the latest at the first interaction. Dottie injects spoken AI disclosure into voice greetings and shows an AI assistant badge on browser voice tests. You must not remove or contradict that disclosure.
- Spoken disclosure: “I’m an AI assistant…” (plus recording notice when storage is on)
- Browser UI: visible “AI assistant” label during tests
- If a caller asks “are you a human?”, the agent is instructed to answer honestly
Section 3
Article 4 — AI literacy for your staff
People who deploy and operate AI systems need enough understanding to use them responsibly. For Dottie that means knowing the agent can be wrong, when to force a human handover, who can hear recordings, and how retention and erasure work.
- Read /ai-literacy with anyone who edits the agent or reviews calls
- Complete the go-live checklist on Dashboard → Voice
- Review sample transcripts after launch and fix knowledge gaps
Section 4
What you must not do
- Market the agent as a human receptionist
- Use Dottie for prohibited AI practices (e.g. social scoring, unlawful emotion recognition)
- Use Dottie for high-risk uses such as employment screening or credit scoring without a separate written compliance programme
- Clone a third party’s voice without rights or use clones for deceptive impersonation
- Treat call quality scores as emotion or biometric analysis of callers — they score agent operational quality only
Section 5
Recordings, transcripts, and privacy
Voice audio and transcripts are personal data. Configure retention under Dashboard → Privacy, restrict who can listen back, and honour export/erase requests. When listen-back or transcripts are stored, Dottie includes a short recording/transcript notice in the greeting.
Section 6
What Dottie is still finishing
Machine-readable in-band marking of synthetic audio (Article 50(2)) is an industry-wide technical track. Dottie already stores structured AI-generated provenance on conversation and recording metadata and enforces spoken disclosure. Trust (/trust) states this honestly — do not claim “full AI Act certification” in your marketing.
This guide is practical operator education, not legal advice. For regulated industries or novel use cases, speak to counsel. Public references: /trust, /aup, /ai-literacy, /dpa.
Mehr Anfragen in gebuchte Aufträge verwandeln
Build agents that answer from your knowledge, qualify leads, and book meetings — on live voice.
Kostenlos starten