BeyondTrust Strengthens Endpoint Security with AI Agent Security, Addressing New Challenges in Privileged Access Management

As organizations increasingly integrate AI coworkers and autonomous agents into their enterprise environments, they encounter a new set of security challenges related to privileged access and endpoint protection. BeyondTrust has responded to these challenges by expanding its Pathfinder platform with the introduction of AI Agent Security, a module designed to enable security teams to monitor and control AI-driven actions in real time.

New Module Enhances Privileged Access Control

BeyondTrust, a recognized leader in privilege-centric identity security, has unveiled AI Agent Security as part of its Pathfinder platform. This new module is engineered to enforce restrictions on what AI coworkers and autonomous agents can execute at the endpoint level, ensuring that actions are regulated before they occur. The module applies consistent privileged-action enforcement across various AI tools currently deployed in enterprises, including Claude Code, Microsoft Copilot, Cursor, and OpenAI Codex.

AI Agent Security marks the first of several planned modules that will be developed natively on the Pathfinder platform. It offers a unified policy framework that integrates with the PathfinderAI and MCP server, extending the AI security capabilities BeyondTrust initially introduced earlier this year at the RSA Conference.

The Challenge of Securing AI Agents

Historically, securing applications on endpoints has revolved around a few key questions: Is the application known to be malicious? Is it behaving suspiciously? Is it authorized to run for a specific user on a particular machine, and what privileges should it have? These questions assume that an application’s behavior can be predicted. However, AI introduces a level of unpredictability with its non-deterministic actions.

AI agents arrive at endpoints with the full privileges of the user who initiated them, operating at machine speed and without adequate safeguards. This lack of oversight allows them to access SaaS, cloud, and production systems, leading to incidents where production environments have been compromised, halting operations. Enterprises are deploying these agents at a pace that outstrips their ability to secure them, with unauthorized AI installations proliferating across enterprise networks.

Marc Maiffret, Chief Technology Officer at BeyondTrust, emphasized the shift in the landscape: “We are not a privileged access company adding AI. 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.”

Centralized Control for AI Tools

AI agents serve as a focal point for privilege management. Their effectiveness increases with the number of systems they connect to and the context they possess. Securing these agents necessitates a singular boundary that is enforced uniformly, regardless of the action’s origin. AI Agent Security provides security teams with a comprehensive approach to monitor, decide, and enforce permissible actions for AI before they are executed. This is achieved through three core capabilities:

• Discovery of AI Assets: The module enables organizations to identify every AI assistant, copilot, and autonomous agent operating across endpoints, both managed and unmanaged. It also surfaces shadow AI instances that employees may have installed independently, mapping their access capabilities.
• Proactive Decision-Making: Organizations can restrict actions to approved AI tools while automatically blocking unauthorized shadow AI. Permissions are granted based on necessity, eliminating the default inheritance of full user credentials. Data boundaries can be established through policy, controlling which MCP servers, plugins, and external services agents may connect to.
• Real-Time Runtime Controls: The module can block unauthorized actions, such as unintended credential exfiltration or the deletion of production code and databases, in real time across various operating systems, including Windows, macOS, Linux, and containers. Consistent controls are applied across all AI tools in the fleet, maintaining a comprehensive, audit-ready record of actions taken, including who performed them and with what authority.

Leveraging Two Decades of Expertise

BeyondTrust has been at the forefront of least-privilege enforcement on endpoints through its Endpoint Privilege Management product. The company is trusted by thousands of clients, including 75 of the Fortune 100, across diverse operating systems. AI Agent Security has been developed in collaboration with Phantom Labs, BeyondTrust’s internal security research team, ensuring that its detection and enforcement capabilities are tailored to the actual behaviors of AI agents rather than assumptions. Recent findings from Phantom Labs indicate that non-human identities now significantly outnumber human ones, with enterprise AI agents experiencing a growth rate exceeding 460% year over year.

Future Availability

AI Agent Security is currently in private beta, with plans for general availability on the BeyondTrust Pathfinder platform set for Autumn 2026. Initially, it will be offered as an add-on to Endpoint Privilege Management.

Source: www.intelligentciso.com

Keep reading for the latest cybersecurity developments, threat intelligence and breaking updates from across the Middle East.