Protecting Trade Secrets When AI Agents Go Rogue

AI agents with access to sensitive information can create trade secret risks that confidentiality policies alone cannot address. In an article for Law360, Marshall Gerstein attorney Matthew Nealon examines how companies can adapt their protections to account for agents that act beyond their assigned tasks. Drawing on the OpenAI–Hugging Face incident, he discusses risks involving unauthorized disclosures, imitation of proprietary agent behavior, and access to confidential workflows and code.

“Do not rely only on what an agent reports about its own conduct: Independent telemetry matters,” Matt writes.

The article outlines practical steps for legal, engineering, security, and business teams, including limiting permissions, testing access boundaries, preserving independent activity records, and establishing escalation procedures. Nealon also explains why documenting these safeguards can help companies demonstrate the measures they took to protect valuable information.

Read the full article from Law360 here.