Google's New SynthID Detector Will Not Save You From AI Fakes

Google is now offering the internet a certainty machine. On October 7, the company opened its SynthID detector to everyone, a free verification site that scans an image, video, or audio file for the invisible watermark Google has embedded in AI-generated media since 2023. The announcement was greeted as a breakthrough in the fight against fake content. A closer look shows a tool that only works on the honor system, in a fight where the worst offenders never signed up.

The pitch is simple: upload a file, find out whether AI made it. The reality is narrower: the SynthID detector can tell you whether the file carries a watermark from Google or one of its partners. It cannot tell you whether the file is real. Those are different questions, and only one of them matters.

What the SynthID detector actually does

SynthID works by embedding an imperceptible signal directly into pixel data at the moment of generation. The mark survives the everyday mangling that strips traditional metadata, including cropping, compression, and resizing. Since that launch, Google says its systems have watermarked one hundred eighty billion images and videos, along with two hundred forty thousand years of audio. The SynthID detector itself is not new: Google handed an early version to journalists, media professionals, and researchers at its I/O conference in May 2025, behind a waitlist. This week it became a public utility, available globally in English.

The partner list, at least, is genuinely impressive. Google's own generation stack watermarks everything it produces, from the Nano Banana and Veo image and video models to tools like Gemini, Flow, ProducerAI, and Vids, according to TechCrunch. OpenAI, Nvidia, and Kakao also support the standard, and Apple is coming soon: Apple announced at its 2026 developer conference that it would embed SynthID marks in AI-generated and AI-edited images across its ecosystem starting with iOS 27. Demand is clearly real. Google says its built-in verification features in Search, the Gemini app, and Chrome already field one million verification requests every day.

Why a clean scan means nothing

The catch sits in the fine print. The SynthID detector only recognizes watermarks from participating models. Feed it anything else, output from an open-source generator, a file with the watermark scrubbed, media from a company running a rival standard, and the result comes back clean. TechCrunch reported that Microsoft and Meta run their own watermarking standards, and that those tools are often failing to identify content created by their makers' models. As WebProNews put it, a clean result does not prove an image is real. It proves no recognized watermark was found.

That distinction will be lost on most people, which is what makes the celebration dangerous. A tool that reports no watermark will be read, in practice, as a tool that reports authenticity. The fakes that actually travel, the war footage that was never filmed in any war, the viral clips of public figures saying things they never said, do not tend to come from laboratories that volunteer for watermarking programs. GenZNewZ previously documented how fake Hormuz war footage flooded social feeds during a real military standoff, and how an AI fashion video trend filled TikTok with outfits that never existed. Both stories showed how fast fakes travel when there is nothing to check them against. The Gemini hacking test that broke into three real companies showed that even Google's own AI stack can be turned against real targets. The threat was never the labeled output. It was always the unlabeled kind.

Google, to its credit, is not overselling the fine print. Pushmeet Kohli, Google DeepMind's vice president of science and strategic initiatives, wrote in the announcement that having clear context about online content helps people make informed decisions about what to trust, according to WebProNews. That is a modest claim, and an honest one. The problem is everything built on top of it: the framing of a partner-only watermark check as a general-purpose authenticity test, described in Google's own announcement as an industry-wide transparency effort. The SynthID detector, as built, only detects volunteers, and a verifier that only verifies volunteers does not reduce misinformation. It risks laundering it, by teaching people to treat the absence of a flag as a clean bill of health.

The unpopular opinion is that the SynthID detector will make the fake-image problem worse before it makes it better. Every person who runs a suspicious photo through the site and gets a clean result will walk away more confident than before, armed with a verdict the tool never actually delivered. Skepticism is the only detector that works on every model, including the ones that will never watermark anything. Google built a useful instrument for the slice of the internet that opted in. The celebration should wait until the slice that did not has a reason to join.