Agentic Systems Add New Layer of AI Hallucination Risk in Legal Work

Unlike traditional generative AI that produces text, agentic systems can take actions based on their outputs, such as sending emails, updating databases, or filing documents, potentially compounding errors if they act on hallucinated information.

This means implementing strict oversight and access controls to ensure that AI actions are reviewed and authorized before taking effect. The article emphasizes that the ability to access multiple software platforms increases both the utility and the risk of AI hallucinations in legal settings.

The article notes that the legal industry has been grappling with AI hallucinations in research and drafting, and that agentic systems represent a significant escalation. Experts quoted in the piece argue that while agentic AI can improve efficiency, firms must update their risk management practices to account for the new failure modes introduced by autonomous actions across platforms.