Looking at user-agents or IPs is the most shallow and non-deterministic way possible. They are arbitrary, I'm not a bot, but I'm using a highly customised one in order to enhance my browsing experience.
>They use genuine IP addresses, user agents, and even simulate mouse movements.
From the list above, only simulating mouse movements part seems like the hardest thing to fake correctly, which genuine IP addresses and user agents is something you can 100% fake. Why focusing on the ip addresses and user agent string then if you can just see that AI Agent is moving it's mouse in a perfect straight line between buttons and doing nothing else with it. Obviously human mouse movement patterns on every webpage are quite chaotic and having it mechanised is an obvious red flag which you should train your model on.
I think the future of ai agent/bot detection is a model trained on user behaviour patters when he is interacting with the page UI.
>They use genuine IP addresses, user agents, and even simulate mouse movements.
From the list above, only simulating mouse movements part seems like the hardest thing to fake correctly, which genuine IP addresses and user agents is something you can 100% fake. Why focusing on the ip addresses and user agent string then if you can just see that AI Agent is moving it's mouse in a perfect straight line between buttons and doing nothing else with it. Obviously human mouse movement patterns on every webpage are quite chaotic and having it mechanised is an obvious red flag which you should train your model on.
I think the future of ai agent/bot detection is a model trained on user behaviour patters when he is interacting with the page UI.