ElevenLabs adds SynthID watermarks to AI-generated audio

ElevenLabs announced that it has started adding Google DeepMind SynthID watermarks to text-to-speech outputs from free users, with broader audio coverage planned over the coming weeks. The company says the marks are inaudible, survive common edits such as trimming and compression, and can be detected through its audio detector. It is a practical provenance move for synthetic audio, especially as voice cloning and deepfake risks keep growing.

Key details: Published June 25, 2026 and updated June 26, 2026, SynthID is being added first to free-user text-to-speech generations, ElevenLabs plans to expand coverage to all generated audio over the coming weeks, The company says the watermarks can be checked with its free audio detector.

Why it matters: Audio provenance is moving from policy talk into default platform infrastructure, which matters for scams, impersonation, and media trust.