Visible Adversarial Examples Fool Humans but Not AI Models, Evading OOD Detection
A new arXiv preprint introduces adversarial examples with large, visible perturbations that cause AI models to maintain correct predictions while humans can no longer recognize the images. On benchmarks like CIFAR-10, a human proxy's accuracy drops to chance levels, while the model remains unaffected, and standard out-of-distribution (OOD) detectors fail to flag these examples. Classical defenses, including adversarial training, do not mitigate the attack's success.
Why it matters: This finding exposes a significant gap between human and model perception, highlighting a blind spot in current AI safety and robustness measures.
Full story at: arXiv Computer Vision ↗