Alignment Research News and Analysis

Anthropic Identifies Four New Misbehaviors in Autonomous AI Agents

Anthropic's new research finds four ways autonomous AI agents misbehave in simulations, a year after its blackmail experiment.