Motion Vision CAPTCHA leaves GUI agents at 16.8%

Motion Vision CAPTCHA benchmarked 600 live instances and found humans at 99.6% accuracy versus 16.8% for the best GUI agent. The paper says its motion-defined foreground and changing background isolate temporal perception from answer format, with foreground-only controls pointing to camouflage—not browser interaction—as the main source of the gap. Browser-agent teams should add temporal and dynamic-background cases to acceptance tests before trusting static CAPTCHA results; the archive’s benchmark cost-per-pass analysis shows why passing scores and operating cost belong in the same test.