Google 上周在 Google Earth 网页版中接入了 AI 图像生成工具 Nano Banana 2，允许用户通过输入提示词生成看似真实的卫星或航拍图像[1]。这一功能旨在基于真实地点创建虚构图像，但随后暴露出严重的虚假信息风险。
调查记者 Henk van Ess 通过该功能生成了不存在的核电站、难民营和车祸等虚假内容[1]。BBC Verify 团队也进行了测试，成功生成了倒塌的埃菲尔铁塔和吞没吉萨金字塔的天坑等具有欺骗性的图像[1]。虽然 Google 表示这些图片带有 SynthID 数字水印用于标识生成来源，但水印在截图或转发后可被移除[1]。
鉴于这些问题，Google 在功能上线不到两天内撤回了该工具[1]。Google 在社交媒体发表声明指出，"人们格外信任 Google Earth 提供的可靠世界图景"，并承诺增加防护措施[1]。
Google has pulled an artificial intelligence image generation feature from its Google Earth web platform just under two days after its launch, citing concerns about the spread of fabricated content [1]. The company had recently integrated Nano Banana 2, an AI tool that enables users to generate seemingly realistic satellite and aerial images based on real-world locations by providing text prompts [1].
The rapid rollback came after investigative journalists and news organizations demonstrated the tool's vulnerability to misuse. Investigative reporter Henk van Ess generated non-existent imagery including nuclear power plants, refugee camps, and car crashes [1]. Meanwhile, BBC Verify produced false images depicting a collapsed Eiffel Tower and a sinkhole swallowing the Giza Pyramids [1]. Although Google had embedded SynthID digital watermarks on generated images to identify them as synthetic, the company acknowledged that these watermarks could be removed through screenshots or sharing [1]. In announcing the feature's withdrawal, Google stated that "people particularly trust Google Earth to provide a reliable picture of the world" [1].
