IA · 7 October 2026 · 4 min read
Google Opens SynthID to the Public: A Single Hub to Detect AI-Generated Media
In brief: Google has launched a public portal for its SynthID technology, allowing anyone to verify whether an image, video, or audio track was generated or altered using artificial intelligence. Moving beyond Mountain View's walled garden, the tool unifies detection for invisible watermarks produced by partners including OpenAI, Nvidia, and Kakao, turning internal provenance tooling into an industry-wide verification standard.
by Team Mocchi's
Determining whether a digital asset is authentic or generated by a synthetic model is no longer restricted to research labs and closed pilot testers. Google has publicly rolled out its SynthID web portal, a globally accessible tool that lets users upload images, video clips, and audio recordings to check for invisible watermarks embedded within pixel and acoustic structures. Previously, verifying content required roundabout prompts through the Gemini assistant or access to an enterprise whitelist; starting today, the process is consolidated into an open web tool.
The update extends beyond a friendlier interface to signal a broader technical alliance. As reported by TechCrunch, the detector processes virtually all widely used digital formats — spanning modern image standards such as AVIF and HEIC to MP4 video streams and uncompressed audio like FLAC and WAV. First introduced in an experimental stage in 2023, Google's watermarking has been baked into models like Veo, Lyria, and the Gemini ecosystem, accounting for over 180 billion processed visual assets and more than 240,000 years of audio tracked at generation time.
A Shared Standard Backed by OpenAI and Nvidia
The crucial strategic shift lies in multi-vendor interoperability. Google's earlier detection tools only scanned for its proprietary variant of SynthID, blind to assets generated by third parties even when those players implemented comparable algorithms. The new release dissolves that silo: the detection engine now recognizes watermark signatures across alliance partners including OpenAI, Nvidia, and Kakao, with Apple support slated to roll out shortly.
Rather than forcing publishers, fact-checkers, and businesses to juggle disparate verification consoles for each commercial foundation model, the site serves as a centralized clearinghouse. While the public tool does not display granular spatial heatmaps detailing manipulated pixels — a feature preserved for internal forensic environments —, it issues an explicit confirmation on whether the file was generated or edited by supported ecosystem tools.
Usage Caps to Defend Against Model Inversion
Opening the platform to the public comes with strict guardrails. Running checks requires signing in through a Google, OpenAI, or Apple account, and users face a ceiling of roughly ten multimedia scans per day. The rationale behind this friction extends beyond server load management.
According to Ars Technica, Google engineers clarified that query rate limiting is essential to thwart adversarial reverse engineering. Unrestricted access would allow attackers to repeatedly query the detector to train automated bypass filters, scrubbing watermark artifacts while retaining visual fidelity. The dynamic cap, which tightens when users test clusters of near-identical files, directly hinders brute-force ablation experiments.
A systemic challenge remains: while SynthID establishes a cohesive front for compliant enterprise model vendors, it cannot detect outputs from unconstrained open-source models hosted locally, nor does it entirely merge with competing provenance frameworks such as the C2PA standard backed by alternative industry coalitions.
Mocchi's take
For European enterprises navigating EU AI Act compliance regarding synthetic content labeling, an interoperable detection portal backed by major model providers is an encouraging step forward. Having a single verification touchpoint streamlines media validation across moderation, publishing, and legal workflows, yet teams must not treat invisible watermarks as a silver bullet. We advise software teams to leverage SynthID checks as an automated first line of defense, backed by cryptographic origin signatures and human oversight wherever data authenticity carries operational or legal risk.