Google opens SynthID Detector to all, limits remain

Google has been developing SynthID since 2023, when it introduced the imperceptible watermark that now underpins the detector. Until now, the company made different versions of its detection tool available to journalists and other media professionals, but availability was severely restricted. It also offered other routes to the same question — the Circle to Search smartphone feature and uploading content to Gemini to ask whether it was created with AI — and the dedicated detector is meant to make those capabilities more accessible.

The tool's usefulness is bounded by the watermark itself. Because it is designed to find hidden SynthID marks, it struggles with AI-generated content that lacks them. Anthropic PBC's Claude and xAI Corp.'s Grok do not use SynthID, and none of the many open-weights Chinese AI models do either, so the detector will miss a great deal of AI-generated material regardless. It also cannot separate fully synthetic images from merely AI-edited ones. Meta Platforms Inc. has experimented with its own watermarking for Muse Image, but that system is useless against images from third-party models.

Google's reach and the fact that several technology heavyweights are already on board may put it in the best position to drive a universal watermarking standard, and the company says it is working on extending partnerships. Whether that happens — and whether it includes the Chinese model makers — is likely to determine how reliable AI-content detection ever becomes.