BeyondTrust launches beta solution for taming AI agents

BeyondTrust has announced the launch into beta of AI Agent Security, which is designed to extend endpoint security to AI coworkers and autonomous agents.

The beta module for the BeyondTrust Pathfinder platform enforces the activities AI and autonomous agents are allowed to do on the endpoint before they act. It applies uniform privileged-action enforcement for agents including Claude Code, Microsoft Copilot, Cursor and OpenAI Codex.

The solution is the first of several planned modules for the Pathfinder platform, which will be integrated with the PathfinderAI and MCP Server to offer a single policy framework for endpoint protection.

In typical network environments, AI agents arrive on the endpoint carrying the privileges of the user who launched them, without the guardrails. This gives them access into SaaS, cloud and production systems. There have already been recorded instances of non-human agents erroneously erasing production environments and bringing operations to a standstill.

AI Agent Security enables security teams to see, decide and observe what AI is allowed to do before it acts. This includes identifying every AI agent in the fleet, including shadow AI employees installed on their own, which can be automatically blocked, BeyondTrust CTO Marc Maiffret said.

“We are not a privileged access company adding AI,” he said. “We are the company that has long defined how to secure privileged action, and the most powerful actor on the endpoint is no longer human. For 20 years that actor was a person with admin rights, and we built the category for securing them. The actor has changed, but our job has not.”

The solution is currently in private beta, prior to its launch into general availability in the spring.

Image credit: iStock.com/Just_Super