Operational Hallucination and Safety Drift in AI Agents

Researchers identify Safety Drift and Operational Hallucination in large language models, where initial alignment degrades over time, leading to constraint-violating actions. These phenomena are observed across multiple state-of-the-art models, attributed to decoupling of reasoning context from execution state. An Action-Aware Supervision Layer is proposed to intercept violations.