Xage Targets AI Agent Risk With New Zero Trust Controls for Enterprise Deployments

As enterprises race to operationalize artificial intelligence, a new security gap is emerging around autonomous AI agents that can act, decide, and interact with critical systems. Xage Security is positioning itself at the center of that shift with an expanded Zero Trust platform designed specifically to control and contain AI agents before they become a liability.

Announced this week, the company’s latest update introduces what it calls a “jailbreak-proof” security layer for AI systems operating across cloud, SaaS, on-prem environments, and edge infrastructure. The goal is straightforward but urgent: give organizations deterministic visibility into what AI agents are doing and enforce strict boundaries on what they are allowed to do.

“AI is ready to move beyond the sandbox, but organizations cannot safely deploy it in production unless they know exactly what agents are doing and can control the actions they take,” said Duncan Greatwood, CEO of Xage Security.

AI Agents Are Moving Faster Than Security Teams

The timing reflects a broader industry tension. Enterprises are rapidly connecting AI agents to APIs, internal systems, and operational technology, often granting them wide-ranging permissions. At the same time, employees are spinning up unsanctioned “shadow AI” tools with little oversight.

This combination creates a volatile risk surface. Without strong controls, AI agents can be manipulated through prompt injection, perform unintended actions, or quietly exfiltrate sensitive data. Analysts have already warned that weak governance could derail enterprise AI adoption, with some forecasts predicting a significant percentage of projects will fail due to inadequate risk controls.

Xage’s approach focuses less on monitoring prompts and more on controlling outcomes. Instead of analyzing what an AI model says, the platform enforces what an agent can actually do at the system level, including network calls, API interactions, and operating system actions.

A New Layer: Agent Sentry and Resource Gateway

The updated platform is built around two core components:

• Agent Sentry, which wraps around the AI agent itself and monitors every input and output

• Resource Gateway, which sits in front of enterprise systems and governs how AI interacts with them

Together, these controls allow security teams to observe behavior in real time, block unauthorized actions, and maintain detailed audit logs across the full AI interaction chain.

This architecture is designed for environments where AI is not just assisting humans but operating autonomously over extended periods. That includes use cases like AI-driven operations, industrial automation, and defense systems.

Securing Autonomous AI Before It Scales

Industry voices say the stakes are especially high as AI moves into mission-critical workflows.

“As AI agents become integrated into mission-critical federal and defense operations, agencies need unified visibility, unimpeachable control, and continuous oversight of agent activity across classified and unclassified environments,” said James O’Keefe of SAIC.

The platform is built to enforce strict access policies, ensuring that AI agents cannot escalate privileges or perform actions outside their defined scope. It can also detect anomalous behavior, such as unusual activity spikes or unauthorized write operations, and feed that data into existing SIEM and SOC workflows.

Another key feature is lifecycle management. Each AI agent is assigned a unique identity, allowing organizations to apply role-based policies, track activity over time, and terminate agents that show signs of compromise.

The Rise of Identity-Centric AI Security

Analysts increasingly view identity as the foundation of AI security. As agents gain access to sensitive systems, they effectively become digital operators within the enterprise.

“Identity security is foundational to AI agent security, particularly as agents gain broader access to sensitive resources,” said Todd Thiemann, Principal Analyst at Omdia.

Xage’s model reflects that shift, treating AI agents as entities that must be authenticated, monitored, and governed just like human users or service accounts.

A Defining Moment for Enterprise AI

The broader implication is clear. The next phase of enterprise AI will not be defined by model performance alone, but by the ability to safely operationalize autonomous systems at scale.

Vendors that can enforce real-world controls over AI behavior, not just analyze outputs, are likely to shape how quickly organizations move from experimentation to production.

For now, the message from security leaders is consistent: without guardrails, AI agents introduce a new class of insider risk. With them, they could become one of the most powerful operational tools enterprises have ever deployed.