Google rolls out improved SynthID AI content detector, now available globally

Google has expanded access to its SynthID content-detection website, enabling users worldwide to check images, video and audio for watermarks from supported AI providers. The tool requires a login (Google, OpenAI, or Apple accounts) and imposes a daily usage quota of about 10 checks per user.

By AI Newsroom· Reviewed by Pranav, Founder & Editor-in-ChiefPublished 6 minutes agoUpdated 6 minutes ago0 views

Google rolls out improved SynthID AI content detector, now available globally

Why It Matters

Making a single public detector available simplifies verification of media produced by participating AI vendors, but the system cannot identify content from providers that use different watermarking, no watermarking, or self-hosted models. That leaves gaps in coverage that limit SynthID’s effectiveness as a comprehensive AI-detection solution.

Key Facts

Feature: SynthID detects watermarks from supported companies in images, video and audio.
Availability: Now available globally via the SynthID website.
Login requirement: Users must sign in with Google, OpenAI, or Apple accounts to use the tool.
Usage quota: Google says each user gets a daily usage quota of approximately 10 image, video, and audio checks.
Purpose of limits: Google engineers told Ars that access limits aim to prevent abuse, such as developing tools to remove SynthID watermarks.

Google has rolled out a globally available version of its SynthID detector, a web tool that checks media for watermarks used by participating AI providers. The site requires users to sign in with an account from Google, OpenAI, or Apple, and Google says each account receives roughly 10 checks per day across images, video and audio. The company has previously applied similar limits and says restricting usage helps deter attempts to reverse-engineer or remove its watermarking.

Despite broader availability, SynthID covers only content that carries the specific watermark signatures implemented by supported vendors. Some companies employ different watermark technologies — Meta was noted as an example — while others publish models without any watermarking. Additionally, open-weight models that users run locally can produce content with no identifying metadata, meaning much AI-generated media will remain invisible to SynthID’s checks.

Google will display a disclaimer on the website acknowledging those coverage limits. The company’s engineering team has also indicated that repeated checks of near-duplicate images can trigger earlier lockouts, a measure intended to flag and block workflows that appear to be probing for watermark removal techniques. Even so, the possibility remains that a method to strip SynthID markers could be developed, which would further reduce the detector’s reliability for older or modified content.

Observers suggest an alternative approach is to cryptographically label authentic media at creation — for example using C2PA standards — so provenance flows with original files rather than relying on post-hoc watermark detection. Google’s Pixel phones are among the devices already employing such provenance labeling. Until wider adoption of cryptographic provenance occurs, SynthID provides a centralized option for identifying AI-origin watermarks from participating vendors, while acknowledging it cannot detect all AI-generated media.