This research examines how personified AI chatbots that adopt human attributes create both trust and dangerous manipulation risks.

Key Findings:

• LLMs increasingly adopt human characteristics (faces, names, personalities) including those of celebrities
• Personification creates false intimacy while increasing user trust
• The EU's AI Act attempts to address these manipulation concerns
• Security frameworks need specific protections against manipulative AI systems

This matters because as AI becomes more human-like, users face increasing vulnerability to deception, manipulation, and potential security threats without appropriate regulatory safeguards.

Manipulation and the AI Act: Large Language Model Chatbots and the Danger of Mirrors