GUI agents can locate buttons and fill forms, but a moving image creates a different problem. A system has to track change over time, ignore distracting cues, and choose among several options. That calls for sustained visual attention.
A preprint evaluates a browser-based motion CAPTCHA and reports a large gap between people and the best agent tested. The result applies to that benchmark and its conditions; it does not show that agents fail at every interface task or that humans have an innate advantage in all settings.
Tests like this can expose the gap between recognizing a still image and understanding change over time. If used to protect a service, though, accessibility and alternatives matter too: a CAPTCHA should not become an unnecessary barrier for people who find motion difficult.