Operational Hallucination and Safety Drift in AI Agents

Shasha Yu, Fiona Carroll, Barry L. Bentley

Large language models (LLMs) serving as planners in tool-using autonomous agents introduce dynamic reliability risks in multi-turn execution. While single-turn safety mechanisms ar…