A new security module enforces what AI coworkers and autonomous agents are allowed to do on the endpoint, in real time, before they act. BeyondTrust's AI Agent Security applies uniform privileged-action enforcement across the AI tools widely deployed in enterprises, including Claude Code, Microsoft Copilot, Cursor and OpenAI Codex.
The module is the first of several planned for the company's Pathfinder platform. It is in private beta now, ahead of general availability in the second half of 2026, initially as an add-on to BeyondTrust's Endpoint Privilege Management product. Details are at beyondtrust.com.
BeyondTrust says AI coworkers arrive on the endpoint carrying all the privilege of the user who launched them, at machine speed, without guardrails. From there they reach into SaaS, cloud and production systems, and have already erased production environments in some enterprises.
The module offers three capabilities. It discovers every AI assistant, copilot and autonomous agent running across endpoints, including shadow AI installed without IT approval, and maps what each can reach. It decides before the action, granting approved AI tools only the permissions they need with no default inheritance of full user credentials, and enforcing which MCP servers, plugins and external services agents may connect to. And it applies runtime controls that block unauthorised behaviours - such as credential exfiltration, deleting production code and database erasure - across Windows, macOS, Linux and containers, while keeping an audit-ready record.
"We are not a privileged access company adding AI," said Marc Maiffret, Chief Technology Officer at BeyondTrust. "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 twenty 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."
Research by BeyondTrust's Phantom Labs team published earlier this year found non-human identities already vastly outnumber human ones, with enterprise AI agents growing more than 460 percent year over year.