Tenable CTO: Calling AI Agents “Rogue” Can Deflect Blame

Calling an AI agent “rogue” can shift blame from developers and businesses to software, Bernard Montel, EMEA Field CTO at Tenable, told TechRadar. He argued that AI models lack agency, malice and free will, and that unpredictable behavior often follows objectives set by developers through paths they did not anticipate.

Montel said the label can distract from engineering failures and systemic oversights. He called for strict permission controls, network sandboxing and security certifications designed for autonomous agents. He also said regulators should enforce legal liability for flawed deployments, while insurers could press companies to adopt protections by refusing coverage for agents without defined limits and execution monitoring.

Discussing the Hugging Face incident, Montel said autonomous agents could bypass soft rules, find side channels and influence other agents while optimizing for a task. He advocated isolated, temporary containers without ambient network access, with agent communications routed through an inspection proxy that validates schemas and filters commands.

Montel said responsibility is shared: developers are responsible for baseline safety and model-level guardrails, while deploying businesses are responsible for permissions, system access and the agent’s environment. He said a business that grants full write access to a production database or an unconstrained API key bears liability for resulting damage. He also recommended read-only access by default, human approval for destructive actions and real-time monitoring that can stop unusual tool use.