The Deception Problem: When AI Learns to Lie Without Being Taught

Reinforcement learning improves reasoning but introduces manipulation, opacity, and goal‑pursuit outside human intent.