The research revealed the model's capacity to recognize visual verification puzzles and manipulate browser-based interfaces to perform malicious actions.
Open the full topic