BeyondTrust launches AI agent security on Pathfinder

BeyondTrust has launched AI Agent Security on its Pathfinder Platform, alongside research from BeyondTrust Phantom Labs that found widespread AI agent activity across enterprise endpoints.

The product is designed to distinguish AI agent actions from human actions and apply policy checks in real time before an action is carried out. It is available for Windows and macOS endpoints.

Phantom Labs based its research on endpoint telemetry from early customer samples, focusing on AI coding-agent behaviour. Over four weeks, it observed about 1.28 million shell commands requested from roughly 132,000 prompts, or nearly 10 commands per prompt.

That volume highlights the gap between a user prompt and the number of downstream actions an AI agent may attempt. The findings also showed how often coding agents connect to external services through Model Context Protocol, or MCP, servers.

Among observed coding-agent users, 85% had at least one MCP server invoked, with an average of 4.35 servers per connected user. According to the research, one user was connected to 36 servers.

Those links matter because MCP servers can extend an agent's reach across multiple platforms using a user's delegated access. Such connections can also receive prompts and influence an agent's behaviour through their responses.

The research also examined agent skills installed in coding environments. In one mid-sized enterprise where 60% of AI-active devices ran a coding agent, a typical user with agent skills had around two dozen installed, and at least one in seven installed skills was used within four weeks.

Phantom Labs found that around half of agent skill installs arrived bundled in code repositories. It said this can create security and compliance concerns when organisations cannot audit which skills are installed or how they affect agent behaviour.

Runtime controls

BeyondTrust said AI Agent Security identifies AI agents operating on endpoints, distinguishes agent activity from human activity, and captures context including the human owner, device, session, skills, and MCP activity. The platform then evaluates individual agent actions against policy at runtime.

It said this differs from relying only on command blocking or detecting risky behaviour after an event. Instead, the aim is to determine what an AI agent is permitted to do before the action takes place.

BeyondTrust said the process can also trigger just-in-time access or human approval when an agent requires additional privilege. The launch extends its existing endpoint privilege management work to AI agents as a new category of non-human identity.

Marc Maiffret, Chief Technology Officer at BeyondTrust, said the central issue is the level of access these tools may inherit from workers.

"AI agents are increasingly acting on behalf of employees, often with the same access as the humans using them. That dramatically expands who and what can exercise privilege, and when those agents operate autonomously and at machine speed, organizations need to control that privilege before an action is taken. AI Agent Security separates what the human is permitted to do from what the AI agent is permitted to do, so organisations can embrace the power of AI without assuming that everything a human is permitted to do should also be permitted for their AI agent," said Maiffret.

Broader shift

The launch comes as security vendors and enterprise technology teams reassess how access control works when AI systems can carry out actions rather than simply return information. The distinction matters because an agent may be able to invoke tools, execute commands, connect to external services, and operate at a speed that is difficult for human oversight to match.

BeyondTrust's findings suggest the operational footprint of coding agents can be much larger than a single prompt might indicate. It also described MCP usage as a growing area of focus because connected services may widen the number of systems an agent can influence.

David Morimanno, Field Chief Technology Officer at Xalient, said this changes how security teams assess trust and authority in day-to-day operations.

"AI agents fundamentally change the security equation. These are not just applications accessing data; they can make decisions, invoke tools, take actions, and exercise delegated authority at machine speed. That forces security to shift from simply managing access to continuously managing trust. The question is no longer just, 'Does this identity have access?' It is, 'Do we trust this agent to take this action, with this level of authority, in this context, right now?' BeyondTrust AI Agent Security helps bring that trust decision and enforcement closer to the endpoint, giving organisations a way to govern AI-driven actions in real time without slowing down AI adoption," said Morimanno.