Google DeepMind has opened its SynthID Detector tool to users worldwide, enabling them to determine whether images, audio, or video were created by artificial intelligence systems from multiple providers including OpenAI, NVIDIA, Kakao, and soon Apple. Previously, people could only check content generated by Google AI by asking Gemini, and verifying material from other AI companies required visiting each provider's separate detection platform. The new consolidated tool marks an expansion from a single-vendor approach to a cross-industry detection system for AI-generated media.
Users can now visit the SynthID.com website and upload image, audio, or video files to have them scanned across all participating partners. The detector provides a yes-or-no answer about whether the content contains an embedded watermark, though Google warns that a negative result doesn't definitively prove the material wasn't AI-generated. The company explicitly states the service "is not a general AI detector" and can only identify media from companies that have implemented SynthID technology. Content created using Meta's or Microsoft's AI systems falls outside the tool's scope because those firms employ different watermarking methods. Heavy editing or modifications can also weaken the watermark signal embedded in the content, leading to false negatives that fail to flag AI-generated material.
According to the report, extensive alterations can degrade SynthID's watermark, requiring users to take "a more comprehensive approach involving additional tools" in such situations. Reddit users quoted in the announcement voiced skepticism about the technology's effectiveness, with one pointing to unacceptably high false positive rates on videos created by the companies' own models. Another user emphasized that defeating the watermark is straightforward, explaining that running a low denoising pass on an image can remove the signal—a technique already testable on OpenAI and Gemini outputs. Despite these limitations, SynthID received recognition on TIME's Best Inventions of 2026 list for its watermarking approach to AI-content detection.
The report discusses C2PA metadata as an alternative strategy that records a file's origin and editing history to establish authenticity. One LinkedIn commentator cited in the piece argues this represents the only "approach that actually scales," characterizing post-creation detection efforts as an arms race. However, C2PA metadata can disappear when content gets uploaded to certain platforms and doesn't transfer to screenshots, limiting its practical durability. SynthID itself functions by embedding an invisible cryptographic watermark directly into content, altering audio frequency probability distributions or pixel and token patterns using a secret key—a technique Google DeepMind launched in 2023 for images, audio, video, and text. The fragmented watermarking landscape across major AI providers means no single solution currently covers all generated content, and the report suggests Google continues working to expand partnerships and broaden detection coverage. For organizations navigating authenticity verification, the choice between embedded watermarks and metadata standards reflects competing bets on whether technical resilience or ecosystem coordination matters more in the long run.