Information Self-Locking: Why RL-Trained AI Agents Stop Asking Questions

Friday, March 13, 2026