K Kulkarni(@ks_kulk):Some great work from the @SuccinctLabs team benchmarking and breaking the leading AI image detectors! Image generation models like Nano Banana and Grok have made it possible to generate photorealistic images at mass scale. Just like adversarial examples became commonplace with the advent of GANs, this study shows that simple edits to AI generated images, such as blurring and noise, can fool even the most sophisticated detectors. This suggests the need for an end-to-end solution to the deepfake problem that doesn't rely on a "cat and mouse" game between attacker and defender. That solution is to use cryptography to establish truth at point of capture. In short, to Prove What's Real.

2026.02.26 20:37

Some great work from the @SuccinctLabs team benchmarking and breaking the leading AI image detectors! Image generation models like Nano Banana and Grok have made it possible to generate photorealistic images at mass scale. Just like adversarial examples became commonplace with the advent of GANs, this study shows that simple edits to AI generated images, such as blurring and noise, can fool even the most sophisticated detectors. This suggests the need for an end-to-end solution to the deepfake problem that doesn't rely on a "cat and mouse" game between attacker and defender. That solution is to use cryptography to establish truth at point of capture. In short, to Prove What's Real.

显示更多