Pangram 推出 AI 图像和视频检测模型 Pangram Image 1.0，可识别 99.2% 的 AI 生成图像，误报率为 0.27%。该模型已集成到 Chrome 扩展中，可在网页任意位置检测 AI 图像和视频，API 也可直接使用；作者称其应对的是 AI 伪造照片用于购物、租房、交友诈骗等现实安全问题。
This startup just dropped the state-of-the-art AI image and video detection model with 99.9%+ accuracy on both.
AI-generated photos and videos are used to scam people online all the time: for shopping, apartments, dating apps, blackmail and wire fraud. Preventing this is one of the biggest emergent problems of AI.
And it's technically very challenging. Malicious actors blur, noise and add artifacts to circumvent detection. Social apps downscale images. Phones do AI-based post processing too. Catching pointed AI edits in a larger real image is tricky, and you need heatmaps to do it.
AI safety on a frontier model level is important, but what's often neglected is the safety problems AI already poses to the world and Im happy someone's working on it.
Plus, this is not a "talk to our sales team" or blog post launch. You can go use the API and Chrome extension today!
Introducing Pangram Image 1.0, our newest model for AI image and video detection. Pangram Image 1.0 identifies 99.2% of AI-generated images while maintaining a 0.27% FPR. It integrates into our Chrome extension, so you can check for AI images and video anywhere on the web.在 X 查看被引用的帖子
来源：Deedy · x.com