Google — the company that buried the Hunter Biden laptop story and built the surveillance architecture that tracks your every click — now wants to decide what's "real" online. Its newly public SynthID Detector, launched globally Wednesday, checks images, video, and audio for invisible watermarks embedded by AI tools. Google frames it as empowerment. It's something else entirely: the censor building the detection gate.
Here's the pitch, as Google DeepMind's Pushmeet Kohli wrote: "As AI tools become better and easier to use, having clear context about the content you encounter online helps you make informed decisions about what to trust." Sounds reasonable. But remember who's talking. This is the same company that cooperated with the intelligence community to suppress a true story days before a presidential election. When Google talks about helping you "decide what to trust," reach for your wallet — and your Bill of Rights.
The mechanics tell the story Google's press release doesn't. SynthID Detector only catches content watermarked by Google and its partners: OpenAI, Nvidia, Kakao, with Apple "soon." Meta uses its own system. Open-weight models — the ones people run on their own hardware — produce content with no watermarks at all. Ars Technica noted that plenty of AI content "will not be identified" and that someone will likely find a way to strip the watermark entirely. Engadget reported the tool can't even distinguish between content fully created by AI and content merely edited with it. So the detector that's supposed to protect you from fakery gives you a fuzzy answer and misses huge swaths of the market.
Then there's the login requirement. You need a Google, OpenAI, or Apple account to use it. Google caps you at roughly 10 checks per day. Google engineers told Ars Technica the limit exists to stop people from developing tools to strip the watermarks. Translation: the gatekeepers are rationing access to the detection tool to protect the watermark system — not to protect you. They're guarding their own infrastructure, and you need an account with one of the club members to get through the door.
Google says Gemini alone has watermarked over 180 billion images and videos and 240,000 years of audio since 2023, with one million verification requests flowing daily through Search, Chrome, and the Gemini app. Those are staggering numbers. They also mean Google is embedding itself as the infrastructure layer for authenticity online — the company that decides which files get flagged and which sail through untouched.
The real stake isn't deepfakes. It's sovereignty over truth. Whoever controls the authenticity standard controls what the public accepts as real. A company that censored a sitting president and a presidential candidate's corruption doesn't get the benefit of the doubt when it builds the machine that labels content "authentic" or not. The tech press — TNW, TechCrunch, The Verge — covered this as a consumer convenience story. Ars Technica and Engadget at least noted the limitations. None of them asked the question that matters: what happens when Google decides the watermark on inconvenient content has been "stripped" — or when content without a watermark gets flagged as suspicious by default?
SynthID isn't a lie detector. It's a loyalty detector — for content produced by the right companies, on the right platforms, under the right terms. The question isn't whether AI detection is needed. It's who runs it, and what they've earned the right to run.
