Google can detect AI watermarks, but news screenshots still require checking against the original source
On October 7, Google announced that SynthID Detector is now available in English to the public worldwide. It can check images, videos, and audio for signs of AI generation. Its key method is identifying invisible watermarks embedded during the generation process.
This tool is worth paying attention to, but “no watermark detected” cannot simply be translated as “this image is real.” Whether a generation tool adds the relevant watermark, and how an image is modified afterward, both affect what evidence can be obtained. Google’s early technical documentation also clearly states that this method is not absolutely reliable in the face of extreme modifications.
There is another distinction: a genuine photograph can still be paired with the wrong time, the wrong location, or a fabricated headline. Identifying how something was made and verifying the news claim made about a photo are different fact-checking tasks.
When I see screenshots of purported official announcements involving $BTC , $ETH , or $SOL , I first find the original statement from the issuing organization, then check the date, scope, and key figures. A screenshot passing some kind of detection test still cannot replace this step, let alone directly trigger a trade.
My view is that AI detection tools can provide additional evidence, but trading processes should still retain the original source. Replacing the entire chain of facts with a single detection result could make a tool intended to reduce false judgments create unwarranted confidence instead. The second image is a reference photo of Google’s office building.
Tap my profile picture to see my live trades
On October 7, Google announced that SynthID Detector is now available in English to the public worldwide. It can check images, videos, and audio for signs of AI generation. Its key method is identifying invisible watermarks embedded during the generation process.
This tool is worth paying attention to, but “no watermark detected” cannot simply be translated as “this image is real.” Whether a generation tool adds the relevant watermark, and how an image is modified afterward, both affect what evidence can be obtained. Google’s early technical documentation also clearly states that this method is not absolutely reliable in the face of extreme modifications.
There is another distinction: a genuine photograph can still be paired with the wrong time, the wrong location, or a fabricated headline. Identifying how something was made and verifying the news claim made about a photo are different fact-checking tasks.
When I see screenshots of purported official announcements involving $BTC , $ETH , or $SOL , I first find the original statement from the issuing organization, then check the date, scope, and key figures. A screenshot passing some kind of detection test still cannot replace this step, let alone directly trigger a trade.
My view is that AI detection tools can provide additional evidence, but trading processes should still retain the original source. Replacing the entire chain of facts with a single detection result could make a tool intended to reduce false judgments create unwarranted confidence instead. The second image is a reference photo of Google’s office building.
Tap my profile picture to see my live trades