EU AI Act — Article 50
The transparency obligation is already in force
Since 2 August 2026, any organisation deploying AI that generates or manipulates audio must disclose it and mark the synthetic content in machine-readable form. This is not a 2027 roadmap item to be budgeted later: it applies to the meetings you are running this quarter.
2 Aug 2026In force. The Digital Omnibus deferred the Annex III high-risk regime to December 2027 — it did not defer Article 50.
€15M or 3%Maximum fine, whichever is higher, calculated on worldwide annual turnover.
Every meetingThe obligation attaches to the deployer, not only the vendor. Your organisation carries it.
Labelling binds the honest. Detection protects you from everyone else.
Article 50 compels legitimate providers to mark synthetic audio. The fraudster cloning your CFO's voice to authorise a transfer will not mark anything. Compliance and security are therefore two distinct problems, and a policy that only addresses the first leaves you exposed to the second.
Kizwork answers both in the same platform: machine-readable marking and explicit speaker consent on every voice we synthesise, and real-time detection with 99.1% detection, alert in 2–3 s for the voices we did not. One is what you show your regulator. The other is what stops the wire transfer.
| What Article 50 requires | How Kizwork delivers it |
|---|
| Disclose that participants are interacting with AI-generated or AI-altered audio | Verified Voice state shown to every participant, in-session, not buried in terms |
| Mark synthetic or manipulated audio in a machine-readable format | Marking applied at generation on speech-to-speech and anonymisation (live in Voice Trust+) |
| Obtain informed consent before reproducing a person's voice | Explicit speaker consent captured before any voice imitation is enabled |
| Be able to evidence the above to a supervisory authority | Forensic reports per incident and immutable audit logs per session |
