AI browsers remain vulnerable to prompt injection attacks via screenshots and hidden content, allowing attackers to exploit users' authenticated sessions.