Google’s deepfake detector system used to debunk McConnell hoax pic

谷歌深度伪造检测系统成功识破麦康奈尔虚假照片

Google’s SynthID system has been used to debunk a high-profile AI-generated hoax image, in a rare but significant win for the system. 谷歌的 SynthID 系统近日成功识破了一张备受关注的 AI 生成虚假图片，这是该系统一次罕见但意义重大的胜利。

Earlier this week, a picture circulated online that seemed to show Kentucky Senator Mitch McConnell covered in tubes in a hospital bed in a state of extreme distress. The image was shared widely on Reddit and X, but by Wednesday, the revered fact-checking site Snopes had debunked the image, noting that, when checked, the image registers as containing the SynthID watermark designed by Google to identify AI-generated pictures. In short, the watermark worked exactly as it was supposed to in a win for anti-deepfake technology. 本周早些时候，一张照片在网上流传，画面显示肯塔基州参议员米奇·麦康奈尔（Mitch McConnell）躺在病床上，身上插满管子，处于极度痛苦的状态。这张图片在 Reddit 和 X 上被广泛传播，但到了周三，权威事实核查网站 Snopes 揭穿了该图片的真面目。Snopes 指出，经检测，该图片包含谷歌为识别 AI 生成图片而设计的 SynthID 水印。简而言之，水印发挥了预期的作用，这是反深度伪造技术的一次胜利。

Senator McConnell’s health has been the subject of intense speculation since he checked into the hospital after an emergency call on June 14. Since that time, he’s been largely absent from the public eye, fueling speculation that his health may be failing. In this case, however, the evidence proved to be entirely fake. 自 6 月 14 日因紧急呼叫入院以来，麦康奈尔参议员的健康状况一直是外界猜测的焦点。此后，他几乎未在公众视野中露面，这加剧了人们对其健康状况恶化的猜测。然而，在这次事件中，证据证明该图片完全是伪造的。

Launched at Google’s I/O developer conference in 2025, SynthID works as an invisible signature, visible to SynthID algorithms but designed to be unnoticeable to the casual observer. Because the signature is built into the image itself, it survives even when an image is screencaptured across multiple platforms, as the McConnell image was. SynthID 于 2025 年在谷歌 I/O 开发者大会上发布，它像是一种隐形签名，SynthID 算法可以识别，但普通观察者却难以察觉。由于该签名直接嵌入在图像本身中，即使图像在多个平台间被截屏传播（如麦康奈尔那张照片一样），签名依然能够保留。

SynthID’s main limitation is that it can only be used when an image-generation tool actively participates in the program. Gemini models have included the watermark since the program launched in 2025. OpenAI joined in May 2026, as part of a broader effort to fight malicious image generation. Anthropic does not participate in the program. Users can check if images contain the watermark by asking a Gemini model or uploading them to OpenAI’s public image verification tool. SynthID 的主要局限性在于，它只能在图像生成工具主动参与该计划时才能使用。自 2025 年该计划启动以来，Gemini 模型就已包含此水印。OpenAI 于 2026 年 5 月加入，作为打击恶意图像生成更广泛努力的一部分。Anthropic 目前未参与该计划。用户可以通过询问 Gemini 模型或将图片上传到 OpenAI 的公共图像验证工具，来检查图片是否包含该水印。