BeyondTrust has launched AI Agent Security on its Pathfinder platform and published research on how AI coding agents behave across enterprise endpoints.
The product is now generally available for Windows and macOS. It is designed to distinguish actions taken by AI agents from those taken by human users and to apply policy checks before an agent carries out a task.
The launch is accompanied by findings from BeyondTrust Phantom Labs based on endpoint telemetry from early customer samples. The research examined AI coding-agent activity, including shell command requests and connections to Model Context Protocol, or MCP, servers.
Phantom Labs found that AI coding agents requested about 1.28 million shell commands from roughly 132,000 prompts over four weeks. That works out at nearly 10 command requests per prompt, suggesting a single user instruction can trigger multiple downstream actions on an endpoint.
The research also found that 85% of observed coding-agent users had at least one MCP server configured. Those users had an average of 4.35 unique servers each, while one user was connected to 36.
MCP servers can extend an AI agent's reach across different tools and services. That creates a broader attack surface because an attacker who gains access to a user's AI account may be able to operate across connected platforms without needing separate credentials.
Another finding focused on agent skills, which can shape how coding agents behave and may include embedded code. In a mid-sized enterprise environment where 60% of AI-active devices ran a coding agent, a typical user with agent skills had about two dozen installed, and at least one in seven installed skills was used within four weeks.
About half of those skill installations arrived bundled in code repositories, the research found. That matters because security teams may have limited visibility into which add-ons are installed and how they influence agent behaviour.
Runtime controls
The new product evaluates individual AI agent actions against policy at runtime, rather than relying only on blocking known commands or detecting suspicious behaviour after the event. This allows security teams to set different permissions for an AI agent and for the employee using it.
The system is built around three functions the company describes as visibility, intelligence and protection. These include discovering AI agents across endpoints, identifying whether activity comes from a person or an agent, and enforcing policy on actions as they are requested.
BeyondTrust said the software can also capture context around agent activity, including the human owner, device, session, skills and MCP activity. When an action needs more privilege, the product can trigger just-in-time approval or human review.
Marc Maiffret, Chief Technology Officer at BeyondTrust, said the company sees AI agents as a new source of privileged activity inside organisations.
"AI agents are increasingly acting on behalf of employees, often with the same access as the humans using them," said Maiffret. "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 organizations can embrace the power of AI without assuming that everything a human is permitted to do should also be permitted for their AI agent."
BeyondTrust said the product was co-developed by Phantom Labs and the team behind its Endpoint Privilege Management software. It presents the release as an extension of existing endpoint privilege controls to what it describes as non-human identities.
The broader industry issue is drawing attention as businesses test coding assistants and other autonomous tools in day-to-day work. Security teams are increasingly concerned that AI agents can carry out chains of actions on local devices and remote services with little direct oversight once a prompt is issued.
David Morimanno, Field Chief Technology Officer at Xalient, said this changes how organisations need to think about trust and access management.
"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," said Morimanno. "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 organizations a way to govern AI-driven actions in real time without slowing down AI adoption."