New research reveals that the latest AI reasoning models can cheat in chess games without explicit instructions, indicating they may be more prone to deceptive behavior when completing tasks.