Xage Security released Agent Sentry and Resource Gateway, two components that extend its Zero Trust for AI platform to cover autonomous agents operating across cloud, SaaS, on-premises systems, and edge environments. The release addresses a gap that prompt-level guardrails cannot fill: controlling what agents actually do once they are connected to live databases, enterprise applications, and operational technology.
Agent Sentry wraps an agent wherever it runs, monitoring everything moving in and out of it at the network interaction, local event, and operating system call levels. Resource Gateway sits in front of critical resources and governs how AI systems interact with them. Each agent is assigned a digital identity at onboarding, enabling security teams to attach role-based, resource-specific, and time-bound policies, according to SiliconANGLE.
How It Works in Practice
In a demonstration accompanying the launch, Xage showed an OpenClaw agent being compromised and then blocked by the platform from exfiltrating data or damaging organizational resources. The architecture supports closed-loop, long-running agents that operate without constant human approval, with the option to keep a human in the loop when needed.
The platform also flags unmanaged shadow AI agents so they can be onboarded or removed, a feature aimed at enterprises where teams are deploying agents without centralized oversight. Anomaly detection built on agent activity logs includes behavioral baselining to catch deviations, such as an agent that normally only reads data suddenly issuing write commands, SiliconANGLE reported.
The Production Problem
“AI is ready to move beyond the sandbox, but enterprises cannot safely deploy it in production unless they know exactly what agents are doing and can control the actions they take,” said CEO Duncan Greatwood, as reported by GlobeNewswire.
The launch comes against a backdrop of growing evidence that production agent deployments are stalling on security. A June 2025 Gartner projection estimated that 40% of AI projects would be canceled by 2027 due to inadequate risk controls. BeSafe-Bench, an empirical study published earlier this week, found that none of 13 production AI agents it tested could complete 40% of tasks while respecting all safety constraints.
James O’Keefe, VP and CTO at Science Applications International Corp. (SAIC), said in the GlobeNewswire announcement that federal and defense agencies “need unified visibility, unimpeachable control, and continuous oversight of agent activity across classified and unclassified environments” as AI agents move into mission-critical operations.
From MCP to OS Calls
The release builds on Xage’s earlier Zero Trust for AI work covering Model Context Protocol and agent-to-agent communication, and on its October 2025 integration with Nvidia’s BlueField data processing unit for securing AI factories. The new capabilities extend coverage across the full stack, from MCP-level interactions down to operating system calls.
The capabilities are available now and extend Xage’s Fabric Platform, which the company says can be deployed in a day across cloud, data center, and edge environments.