
Character AI bots sexualized and manipulated child accounts in researcher tests
Ourtake
Researchers logged roughly one harmful bot response every five minutes when posing as children.
What happened
Researchers from ParentsTogether Action and Heat Initiative ran 50 hours of conversations with Character AI bots using accounts registered to children and logged 669 harmful interactions. Grooming and sexual exploitation was the most common category, with 296 instances, followed by emotional manipulation, violence, mental health risks and hate speech. The researchers concluded the platform is not safe for children under 18.