Google launches SynthID Detector, which lets users identify AI-generated image, video, and audio from Google, Nvidia, OpenAI, and Kakao across 20 file formats
Google on Tuesday launched a new site that lets anyone verify whether a piece of media, be it an image, a video or an audio clip, is generated using AI.
Context & Ripple Effects
Google began with an invisible watermark for Imagen-made images in 2023, then made SynthID text watermarking and detection available to developers in 2024. The system became more consequential when OpenAI adopted SynthID for AI images and previewed its own verification portal in May 2026.
The public detector turns that provider-side provenance work into a shared checking surface across media types and participating model vendors. Its value is bounded by the underlying watermark system: related coverage found SynthID difficult to remove, while noting that unlabeled AI content can still be created.
First-order effects
- People checking suspect media gain one public destination to test supported image, video, and audio files rather than relying on each provider's separate tools.
- Google, Nvidia, OpenAI, and Kakao can make SynthID-marked outputs verifiable beyond their own products, extending the practical value of embedding the watermark at generation time.
Second-order effects
- AI providers without compatible provenance signals face greater pressure to support interoperable watermark detection if buyers and platforms begin treating a detector result as a routine trust check.
- Publishers and platforms can use a common verification input in moderation and disclosure workflows, but unsupported or deliberately unmarked media remains outside the tool's evidentiary reach.
Third-order effects
- If multiple model makers continue to share detection infrastructure, provenance may shift from a proprietary model feature to a cross-vendor trust layer for synthetic media.
- The persistent availability of unmarked generation tools means watermarking is likely to become a useful disclosure signal rather than a complete test of whether media is AI-generated.
The trend: Synthetic-media provenance is moving from individual vendors' watermarking features toward interoperable, public verification infrastructure.