Google's SynthID detector goes global in English; Apple Intelligence watermarking to follow 'soon'

Google has upgraded its SynthID AI content detector from limited early access to a global release, available in English, according to Google's announcement as reported by 9to5Google on October 7, 2026. The tool checks images and videos for SynthID's invisible watermarks. It is no longer limited to content from Google's own models and its initial partners.

What changed

• Availability: SynthID moves from limited early access to global availability. Google says the tool is English-only for now and is expected to expand further.
• Partner coverage: Detection now works with content from Google's Gemini models and from partners including OpenAI, Nvidia, Kakao and Apple.
• Apple timing: Google says images made with Apple Intelligence will carry a SynthID watermark "soon." No date has been given, so detection of Apple-generated content is not yet reliable.

How SynthID works

SynthID embeds invisible watermarks at the generation stage. The detector later scans an image for those watermarks to indicate whether an algorithm produced it.

This has a hard limit: detection only works if the generating tool inserts a SynthID watermark in the first place. Content from a model or service that does not embed one will not be flagged, and a clean result is not proof that an image is human-made. The partner list defines the coverage.

What is not disclosed

The source report does not include:

• Detection accuracy, false-positive or false-negative rates
• Robustness of the watermark to cropping, compression, screenshots or editing
• Pricing or usage limits for the detector (pricing not yet disclosed)
• Which surfaces the detector is available on, or a timeline for languages beyond English
• Whether partners' existing models already embed SynthID watermarks, or only future outputs will

The claims here come from Google's press release. We have not seen independent testing of the detector.

What this means

The partner list is the main story. A watermark detector is only as useful as the share of generated content that carries the watermark. Adding OpenAI, Nvidia, Kakao and Apple turns SynthID from a Google-only check into a candidate cross-vendor verification layer, which makes it more useful to journalists, platforms and ordinary users.

The gaps matter just as much. Open-weight image and video models run locally, and any generator outside the partner list produces content without the watermark. A negative detection result will stay ambiguous. Apple's "soon" also leaves a gap for images made on a large installed base of devices.

For builders, the practical question is whether Google exposes SynthID detection as an API or only as a consumer tool, and whether watermark robustness holds up under real-world editing. Neither is answered in this announcement. Until independent evaluations appear, treat SynthID as one signal of AI provenance, not a definitive authenticity check.