Google makes SynthID Detector open to all and extends verification to OpenAI and NVIDIA content

Google is opening SynthID Detector to everyone to verify images, videos, and audio files flagged by Google, OpenAI, NVIDIA, or Kakao.

Google DeepMind is now opening SynthID Detector to the general public. Since October 7, 2026, the tool has been accessible globally in English, allowing users to analyze an image, video, or audio file to check for the presence of a SynthID watermark. The technology is no longer limited to content produced with Google's tools: the detector now also recognizes content from partners like OpenAI, NVIDIA, and Kakao, while Apple is expected to join this ecosystem soon.

SynthID has existed since 2023 and is based on a different logic than traditional AI-generated content detectors. Rather than looking after the fact for characteristics supposed to betray artificial generation, the technology adds an imperceptible digital signal directly to the content at the moment of its creation. The detector then specifically searches for this signature. Google DeepMind details how this approach works on the official SynthID page.

This distinction is important. A negative result does not mean that content is authentic or that it was not created with AI. It only indicates that SynthID did not find a sufficiently reliable signal. A file coming from a model that does not use this technology can therefore go completely undetected, as can content whose signature has been sufficiently degraded by successive transformations. Google itself recommends not considering SynthID as sole proof.

Conversely, when the watermark is detected, the result indicates that the media was likely generated or modified by a system using SynthID. The technology is designed to resist common transformations such as compression, certain color changes, or various editing operations, without being presented as infallible. Google notably acknowledges that significant transformations can eventually deteriorate the signal.

The public portal currently analyzes three categories of media: images, videos, and audio. For a video or an audio track, SynthID can also identify the portions in which the signal is detected. The technology itself can also be applied to text, but this modality is not offered in the current public version of the SynthID Detector.

The change in scale is notable. Google indicates that it has already applied SynthID to more than 180 billion images and videos since its launch, as well as to the equivalent of 240,000 years of audio content. The group had opened an initial version of the Detector in 2025 for journalists and media professionals before extending access to everyone today.

The interest of the launch also lies in the progressive opening of SynthID beyond the Google ecosystem. OpenAI, NVIDIA, and Kakao now use the technology on certain content generated by their services. Google also announces Apple as an upcoming partner. The Detector is thus gradually becoming a common verification point for several players rather than a tool dedicated solely to Gemini and Google models.

However, this approach does not replace other provenance mechanisms. In parallel, Google continues to develop support for standards like Content Credentials, while SynthID provides a different layer by embedding its signal directly into the content. The goal is more to cross-reference multiple clues than to base the authenticity of a media on a single verdict.

Moreover, Google already integrates SynthID verification into several of its products. Similar checks are available in Gemini, Google Search, and Chrome. According to the company, these verification features now process more than one million requests per day.

The operation of the portal remains deliberately framed. Google notably prohibits its automation, its use as a bypassed API, or any attempt to multiply transformations of content to understand how to bypass the detector. According to the terms of service, files sent for analysis are deleted once the result is provided. However, a digital signature of the file may be kept for 24 hours to manage quotas and prevent abuse.

With the opening of SynthID Detector and the arrival of external partners, Google is primarily trying to transform its in-house watermark into a more widely shared provenance infrastructure. The service does not answer the question "is this content AI-generated?" in absolute terms. It answers a more precise, but potentially more reliable question: "does this content bear the signature of a model that has adopted SynthID?"