Policy & Safety

Watermark rules go global as voice AI crosses the 'indistinguishable' threshold

With synthetic voices now passing blind listening tests, a coalition of regulators moved in concert to mandate provenance labels on AI-generated audio and video.

By Sofia Lindgren, AI Policy Correspondent — GENEVA

GENEVA — Regulators across four major jurisdictions moved in concert this week to require provenance labels on AI-generated audio and video, days after a peer-reviewed study confirmed that leading synthetic voices now pass blind listening tests against genuine recordings.

The coordinated rules mandate cryptographic content credentials at the point of generation and require major platforms to surface them to users, with fines scaled to global revenue for systematic non-compliance.

Civil-liberties groups broadly welcomed the provenance approach — which labels synthetic content rather than banning tools — while warning that enforcement against offshore and open-source generators remains the hard, unsolved half of the problem.

Enable JavaScript to read the full story on Neural Daily News.