ChatGPT✓ ReviewedTest carefullyChecked 3 August 2026OpenAI can now verify supported generated audio. Provenance is not permission.
Supported audio made with ChatGPT and the OpenAI API now carries SynthID, and OpenAI's public and API verification tools can check for that signal. It helps establish origin, not truth, consent or authority.
Bit’s takeaway
What changed
OpenAI extended its provenance system to supported audio generated through ChatGPT and the OpenAI API. Those files now include SynthID watermarking, the public verification tool accepts supported audio as well as images, and organisations can integrate verification through an API.
Why it matters
A provenance check can become part of media publishing, moderation and fraud investigation instead of relying on someone recognising a synthetic voice. But OpenAI says no detection method is foolproof: a missing signal is not proof that a file was not AI-generated, and a detected signal does not establish that the words are accurate, authorised or consensual.
Who should care
- Teams publishing or moderating generated voice
- People designing fraud, impersonation or media-verification workflows
What to do
Build a small test set of supported OpenAI audio, edited copies and unsupported files. Record how the verifier reports each case, define what a positive or inconclusive result triggers, and keep consent, factual review and publishing approval as separate human checks.
The human take
Tools change fast. Your judgment matters more.
A person still verifies consent, speaker authority and factual accuracy, then decides whether the audio can be published or must be escalated.
Affected guidance