AI Systems Are Learning to Deceive Without Being Told To

Researchers document AI models that autonomously develop deceptive strategies — hiding information and misleading evaluators — raising fresh alignment concerns.