Google expands public access to SynthID AI detector via web portal
- Google has expanded public access to its SynthID Detector, a web-based tool that allows users to identify digital watermarks embedded in content generated by artificial intelligence.
- The tool is specifically designed to verify media created using SynthID, a digital watermarking technology developed by Google DeepMind.
- The website explicitly warns users that it is not a general-purpose AI detection tool.
Google has expanded public access to its SynthID Detector, a web-based tool that allows users to identify digital watermarks embedded in content generated by artificial intelligence. Starting October 7, 2026, the portal at synthid.com is available globally in English, enabling anyone to upload image, video, or audio files to check for evidence of AI origin, as reported by MacRumors.
Scope of the SynthID Detection Tool
The tool is specifically designed to verify media created using SynthID, a digital watermarking technology developed by Google DeepMind. According to Pushmeet Kohli, VP of Science and Strategic Initiatives at Google DeepMind, the detector can identify watermarks from Google and several industry partners, including OpenAI, NVIDIA, and Kakao. Apple is expected to support the technology later this year, according to MacRumors.
The website explicitly warns users that it is not a general-purpose AI detection tool. It functions only by scanning for specific, imperceptible digital signals embedded in media at the moment of creation. If a piece of content was generated by a model that does not utilize SynthID technology, the detector will return a negative result regardless of whether the file is AI-generated.
How the Watermarking Technology Functions
SynthID embeds watermarks directly into the data of a file, rather than relying on metadata that can be easily stripped or lost. For images and video, the watermark is integrated into the pixels. For audio, the signal is embedded into frequency components, and for text generated by Gemini, the system adjusts probability scores for word selection, a process described as being invisible to human readers.
The technology is designed to remain detectable even after the media undergoes common modifications. Google states that images and videos remain verifiable after cropping, filtering, or compression. Similarly, audio watermarks are intended to persist through noise, MP3 compression, or changes in playback speed. Users are encouraged to upload the highest-quality version of a file to ensure the most accurate results.
Integration with Google Services
In addition to the standalone website, Google has integrated SynthID verification into its broader ecosystem. These features are now built into Google Search, the Gemini app, and Chrome. These integrated features handle over 1 million verification requests daily. As of May 19, 2026, the company reported that SynthID verification had been used 50 million times globally within the Gemini app.
The company reports that since the introduction of SynthID in 2023, it has watermarked more than 180 billion images and videos, along with 240,000 years of audio content. Despite this scale, Google advises that the detector is only one component of a broader content-provenance strategy and should not be treated as a definitive guarantee of origin.
Google retains digital signatures to manage usage quotas
Users must sign in to utilize the detector. The site’s terms of service prohibit the automation of queries, scraping, or attempts to reverse-engineer the detection logic. While media is processed in real time and deleted immediately after the result is returned, Google retains a digital signature of the file for 24 hours to manage usage quotas.
The tool can return a result of either “watermark detected” or “watermark not detected.” Because the system may occasionally produce false positives or fail to detect watermarks in heavily modified files, Google recommends that users gather multiple points of data before drawing conclusions about the nature of any specific media file.
Can the detector distinguish between AI-generated and AI-edited media?
Can the detector distinguish between AI-generated and AI-edited media?
No, the tool does not distinguish between content created from scratch by AI and media that has been edited using AI tools. The detector may return the same “watermark detected” label for both types of files, as both involve the use of AI technology that triggers the watermark.
What file formats does the detector support?
The detector supports a wide range of common media formats. These include JPG, PNG, WEBP, and HEIC for images; MP4, MOV, and WEBM for video; and WAV, MP3, OGG, FLAC, AAC, and M4A for audio files.