Xage Unveils 'Jailbreak-Proof' Security to Tame Autonomous AI Agents

PALO ALTO, CA – May 27, 2026 – As artificial intelligence rapidly evolves from experimental sandbox tool to autonomous operator within corporate and government networks, the risk of rogue behavior, data theft, and catastrophic error has become a primary barrier to adoption. Addressing this challenge head-on, cybersecurity firm Xage Security today announced a major expansion of its Zero Trust platform, designed to provide a “jailbreak-proof” security layer for autonomous AI agents.

The new capabilities promise to give organizations deterministic, end-to-end visibility and control over AI agent actions across cloud, SaaS, data center, and edge environments. The move comes as businesses and government agencies grapple with the urgent need to operationalize AI for a competitive advantage while simultaneously protecting their most critical assets from these powerful new non-human actors.

“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, in the company’s announcement. “Xage provides the deterministic visibility and enforcement needed to prevent rogue behavior, manipulation and unintended consequences.”

The AI Security Imperative

The rush to integrate AI has created a new and dangerous class of insider threat. AI agents, often connected to APIs, databases, and internal applications, are being deployed with alarming speed. In many cases, employees are deploying their own “shadow AI” agents, granting them broad, unmonitored access to sensitive resources. This rapid, often uncontrolled, proliferation has left a massive security gap that traditional tools are ill-equipped to handle.

Industry data highlights the growing sense of urgency. A recent Check Point report found that 78% of organizations have already experienced AI-related security incidents. This reality underpins Gartner's prediction that by 2027, 40% of AI projects could be canceled due to inadequate risk controls. The primary threats are no longer theoretical; they include prompt injection attacks, where malicious instructions hidden in documents or data cause an agent to act against its programming, and AI automation hijacking, where an attacker seizes control of an autonomous system.

“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,” noted James O’Keefe, a strategist at SAIC. “Secure governance of AI agents will be essential to scaling AI adoption while maintaining mission integrity and resilience.”

A 'Jailbreak-Proof' Architecture

Xage’s claim of being “jailbreak-proof” hinges on its unique architectural approach. Instead of focusing on filtering prompts or analyzing model outputs—methods that have proven brittle—the platform enforces policy at a deeper level. The solution combines two core components: the Xage Agent Sentry and the Xage Resource Gateway.

The Agent Sentry acts as a protective capsule around the AI agent, monitoring every interaction at the network and operating system level. The Resource Gateway stands guard in front of critical resources—databases, applications, or control systems—and governs how AI agents can interact with them. Together, they create a system where policy is enforced at the protocol layer, below the AI itself. This means that even if an agent is successfully tricked by a malicious prompt, its attempts to perform unauthorized actions, like exfiltrating data or running a harmful script, are blocked before they can be executed.

To demonstrate this, Xage showed its platform blocking a compromised OpenClaw agent from causing harm. This approach of controlling what an agent does rather than what it says aligns with a growing consensus among security experts. As one analyst from Omdia, Todd Thiemann, stated, “Identity security is foundational to AI agent security... Organizations need to gain visibility to their entire AI agent estate, enforce granular policies around what agents are permitted to access, ensure AI agent identity governance, and establish lifecycle management.”

To handle the immense scale of modern AI deployments, Xage has also collaborated with NVIDIA to integrate its security fabric with BlueField Data Processing Units (DPUs). This provides hardware-accelerated enforcement capable of supporting up to 10 million assets and 50 million simultaneous interactions, ensuring security doesn't become a bottleneck for performance in high-stakes AI factories and critical infrastructure.

Governing the Autonomous Frontier

The implications of this technology extend beyond simple threat prevention. By providing a reliable safety net, Xage aims to unlock the use of AI in closed-loop, autonomous applications where agents operate for long periods without direct human supervision. This is crucial for industries like manufacturing, energy, and logistics, where AI can optimize complex processes in real time.

“AI agents are rapidly becoming autonomous operators inside enterprise and government environments, and the ability to monitor, contain and control those agents will define the next era of operational advantage,” said Joe Besselman, a former U.S. Air Force program director. “Organizations that can observe agent behavior, block risky actions, and maintain trusted audit trails will be the ones that can effectively operationalize and secure AI successfully.”

Xage’s platform assigns each AI agent a unique, cryptographically-secured digital identity, allowing security teams to apply granular, role-based access controls. This identity-first approach is also used to detect and manage previously unknown “shadow AI” agents operating on the network. By baselining normal behavior, the system can detect anomalies—such as an agent that typically only reads data suddenly attempting to write—and flag them for review or block them automatically, feeding detailed logs into existing Security Information and Event Management (SIEM) systems for comprehensive oversight.

A Crowded Field and Lingering Challenges

Xage is entering a fiercely competitive and rapidly expanding market. Established cybersecurity giants like Palo Alto Networks and Microsoft, along with a host of specialized startups such as Lakera Guard and HiddenLayer, are all racing to offer solutions for securing AI workloads. The global market for AI governance is projected to surge from around $300 million in 2025 to nearly $6 billion by 2035, signaling a massive investment wave in securing artificial intelligence.

Despite these advancements, significant challenges remain. The threat landscape is evolving at a breakneck pace, with adversaries already using AI to craft more sophisticated phishing attacks and malware. Integrating any new security platform into a complex and often fragmented corporate IT environment presents its own set of hurdles. Furthermore, the effectiveness of any AI-driven security system depends heavily on the quality and consistency of the data it ingests, with data poisoning and inherent model biases remaining persistent concerns.

Ultimately, the technology itself is only one part of the solution. A profound shortage of professionals skilled in both AI and cybersecurity threatens to slow the adoption of these vital tools. For organizations racing to deploy AI, the ability to prove these complex, autonomous systems are truly under control may ultimately determine the winners and losers in the next technological revolution.