OpenAI Adds SynthID Watermarks to GPT-Live Audio — and Ships a Developer Verification API

Starting July 31, all GPT-Live audio via ChatGPT Voice and the OpenAI API carries Google's SynthID watermark baked into the signal — and a new verification API lets developers check AI provenance inside their own pipelines.

Audio generated through GPT-Live now carries a dual provenance stack: C2PA metadata (editable, but auditable) and a SynthID watermark embedded in the audio signal itself — robust to cropping, compression, and format conversion. OpenAI's public verification portal detects both signals.

The developer-facing piece is the new verification API endpoint: incorporate provenance checks into content moderation pipelines, media publishing platforms, or any system that needs a programmatic "was this AI-generated?" signal on audio.

Why it matters: GPT-Live now powers audio generation in millions of applications. The watermark travels with the content, not just the strippable metadata — and the verification API means you can add that check to your own pipeline without building audio-fingerprinting detection yourself.

Sources: Advancing content provenance — OpenAI · Introducing GPT-Live — OpenAI · SynthID — Google DeepMind · C2PA