Google Unveils SynthID Detector for AI Media

Google has made its SynthID Detector publicly available, enabling users to identify AI-generated media. Launched in 2023, SynthID embeds an invisible watermark into content created by Google's AI models. Previously accessible only to select researchers and journalists, the detector now allows general users to upload files and receive a yes/no indication of an AI watermark.

The SynthID Detector scans for watermarks embedded by Gemini and other SynthID partners, including OpenAI, NVIDIA, and Kakao. This contrasts with earlier methods where users had to ask Gemini directly about content authenticity, a process that lacked independent verification. The detector's simplicity, focusing on watermark detection, aims to eliminate false positives but means it cannot identify AI content from models that do not use SynthID, such as Meta's tools.

Google states that SynthID watermarks are designed to persist even after media editing. To prevent misuse and attempts to bypass the system, users are limited to approximately 10 checks per day and must be logged into a Google, OpenAI, or Apple account. While other companies like OpenAI and Meta also offer AI detection tools, they often rely on specific markers, and no universal detector currently exists that can identify all forms of generative AI content.