Xage Security announced major enhancements to its Zero Trust for Artificial Intelligence (AI) platform, providing a jailbreak-proof security foundation for autonomous AI agents in closed-loop and high-stakes applications. The new AI security capabilities give enterprises complete visibility into AI interactions and precise control over AI agent behavior across distributed and hybrid environments. They provide deterministic visibility and control over AI agents, enabling secure production deployments across SaaS, cloud, in-house data center and edge.
As part of a platform demonstration, Xage showed how an OpenClaw agent could be hacked and manipulated, and how its updated Zero Trust for AI platform blocked the compromised agent from damaging critical organizational resources or extracting data.
Announced on Wednesday, the expanded Zero Trust for AI platform introduces new protections aimed at creating a jailbreak-resistant security foundation for autonomous AI agents operating in closed-loop and high-stakes environments. Xage said the capabilities provide enterprises with detailed visibility into AI interactions and granular control over agent behavior across distributed and hybrid infrastructures.
“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 Duncan Greatwood, CEO of Xage Security. “Xage provides the deterministic visibility and enforcement organizations need to prevent rogue behavior, manipulation and unintended consequences. With Xage, enterprises can confidently put AI’s potential into action across high-stakes real-world environments, from cloud and SaaS applications to on-prem and edge systems.”
Enterprises are rapidly moving AI agents closer to production as they connect them to APIs, SaaS platforms, databases, internal applications, cloud services and operational technology (OT) environments. Meanwhile, individual users are deploying their own ‘shadow AI’ agents, often granting them broad access to critical resources.
Many organizations lack deterministic visibility and controls needed to govern what these agents can see, do, and change. Without strong enforcement, agents may be manipulated by prompt injection, take unauthorized actions or exfiltrate sensitive data. Although Gartner previously predicted that 40% of AI projects would be canceled by 2027 due to inadequate risk controls, Xage enables enterprises to move AI from sandboxed experimentation into real-world production environments with confidence.
Xage delivers end-to-end visibility and control across the full AI interaction chain, including users, agents, LLMs, tools and cloud or internal applications. Its new Zero Trust for AI solution combines two core capabilities. Xage Agent Sentry encapsulates AI agents wherever they operate and monitors all inputs and outputs associated with the agent. Xage Resource Gateway sits in front of critical resources and governs how AI systems interact with them. Together, these capabilities allow organizations to see exactly what agents are doing, block unauthorized behavior and maintain detailed logs for governance and audit. Unlike solutions focused on prompts or model outputs, Xage controls the actions agents can actually take at the network-interaction, local event and OS-call levels.
To move AI beyond constrained pilots, organizations must address the practical risks of agency.
Xage said its architecture is designed to address several critical production scenarios involving autonomous AI systems. The platform can govern access to sensitive enterprise data by allowing AI chatbots to read specific database records while blocking unauthorized modifications. Its multihop capability is intended to prevent privilege escalation when a low-privileged user interacts with a highly privileged AI system.
The company also said the platform can stop prompt injection attacks and rogue AI behavior. If an AI agent encounters hidden malicious instructions within a document and attempts to generate a harmful script or perform an unauthorized action, Xage Agent Sentry is designed to detect and block the activity.
For closed-loop autonomous AI systems operating for extended periods without constant human oversight, Xage said its platform enforces security policies and helps limit unintended consequences as agents make changes and adapt based on feedback. Organizations can configure deployments for fully autonomous operation or maintain human oversight within the decision-making process.
Xage provides a practical foundation for managing AI agents throughout their operational life. Each agent is assigned a secure digital identity upon onboarding, allowing teams to define agent-specific policies based on role, resource and time-bound need. Xage even detects unmanaged or ‘shadow AI’ agents, so that they can either be onboarded for management or removed.
If an agent is compromised, Xage blocks its attempts at harmful actions, limiting the blast radius of the attack. By recording detailed information about AI agent actions, Xage Security said its platform enables advanced anomaly detection capabilities. These include behavioral baselining to identify deviations such as unusually high activity levels or unauthorized write actions from agents that typically only perform read operations.
The platform is also designed to function as an early warning system by flagging unexpected or potentially risky behavior for review before it escalates into a larger security issue. In addition, Xage said logs and detected anomalies can be integrated into existing SIEM and SOC tools to support monitoring and response across large-scale enterprise deployments.
This announcement builds on Xage’s previously announced Zero Trust for AI capabilities for MCP and A2A. Xage is now providing comprehensive protection against AI abuse for all of an organization’s critical resources, including MCP- and API-accessible assets, SaaS applications, cloud services and on-prem and edge systems. By securing both the agent itself through Agent Sentry and the resources it touches via the Resource Gateway, Xage wraps AI activity with jailbreak-proof visibility and control.
