Gartner has previously indicated that 40 percent of AI projects could be cancelled by 2027 due to inadequate risk controls. That projection is less a forecast and more a diagnosis of the governance gap that most enterprises are currently carrying in silence.

Xage Security’s announcement of major enhancements to its Zero Trust for AI platform is a direct response to that gap, and the architectural choices the company has made reveal something important about where enterprise AI security is actually headed.

What Xage Has Built and Why the Architecture Matters

The Xage Zero Trust for AI platform is organized around two interlocking capabilities: Agent Sentry and Resource Gateway.

Agent Sentry encapsulates the AI agent wherever it runs, creating a monitoring envelope around everything entering and exiting the agent. This is not prompt-level filtering or output scanning. It operates at the network-interaction, local event, and OS-call levels, which means it captures what the agent is actually doing in infrastructure terms rather than what it claims to be doing in language model terms. That distinction is architecturally significant.

Resource Gateway sits in front of critical systems to govern how AI agents interact with them.

The two capabilities together create a chain of custody across the full AI interaction sequence: users, agents, large language models, tools, and both cloud and on-premises applications fall within the visibility and enforcement perimeter.

The practical implications are substantial.

An AI agent that receives a document containing hidden malicious instructions and attempts to execute an unauthorized script can be detected and blocked at the action level, not just flagged after the fact. A low-privileged user who connects to a highly privileged AI agent does not inherit elevated access through that chain. An agent operating in a closed-loop environment over extended periods without constant human review is still governed by policy enforcement that limits the blast radius of any compromise.

Each AI agent is assigned a secure digital identity on onboarding, enabling role-based, resource-specific, time-bound policy definition. The platform also surfaces unmanaged shadow AI agents, the independent deployments that individual users or teams have connected to enterprise resources without formal governance, allowing security teams to either bring them into the management framework or remove them.

Shadow AI Is the Insider Threat Problem Nobody Has Named Yet

The shadow AI dimension of this announcement deserves particular attention from enterprise security architects, because it maps onto a risk category that is currently being systematically underreported.

Shadow IT has been a persistent governance challenge for two decades. The shadow AI version of that problem is structurally more dangerous because the entities involved are not simply storing data in unapproved locations. They are taking actions. An employee who configures a personal AI agent with access to internal applications, a shared database, or a customer-facing API is not just introducing a data sovereignty risk. They are introducing an autonomous actor with potentially broad access rights that has never been reviewed, never been scoped, and never been logged.

Security operations teams have limited visibility into these deployments under current architectures because the agents present as authorized user sessions, not as separate entities. Detecting them requires the kind of behavioral baselining and identity-layer visibility that Xage’s platform is specifically designed to provide.

For CISOs managing AI governance policy, the practical question is not whether shadow AI agents exist in their environment. They almost certainly do. The question is whether the current architecture has any mechanism to find them, scope them, and apply controls before one of them does something that becomes a regulatory or reputational event.

Federal and Defense Signals With Direct Commercial Implications

Two of the endorsements in Xage’s announcement carry strategic weight beyond their immediate context.

James O’Keefe of SAIC, speaking specifically about federal and Defense AI deployment, frames the requirement as unified visibility and unimpeachable control across both classified and unclassified environments. That language reflects a procurement posture that is increasingly shaping commercial enterprise requirements as well. Regulated industries, critical infrastructure operators, and large financial institutions face analogous governance requirements: AI agent activity must be auditable, containable, and demonstrably compliant with access control policy.

Joe Besselman, with a background that includes being Program Director for Global Combat Support Systems at the US Air Force, frames the competitive dimension clearly: organizations that can observe agent behavior, block risky actions, and maintain trusted audit trails will have an operational advantage over those that cannot. In commercial terms, that translates directly into the difference between enterprises that can accelerate AI deployment with governance confidence and those that remain in perpetual pilot mode because they cannot satisfy their own risk committees.

The federal and defense interest in this capability class is a leading indicator. That sector’s procurement requirements consistently shape what enterprise security buyers prioritize in subsequent cycles.

Market Signals and Buyer Intent in the AI Security Category

The AI security market is currently fragmented across several overlapping sub-categories: model security, prompt injection Defense, data loss prevention for AI outputs, AI access management, and agent governance. Most enterprises encountering this space for the first time are attempting to map existing security frameworks onto a problem that those frameworks were not designed to address.

Xage’s positioning is notable because it approaches the problem from an identity and access enforcement perspective rather than a content filtering perspective. That is a meaningful architectural choice. Content-level filtering is inherently probabilistic and model-dependent. Access-level enforcement is deterministic.

For enterprises that need to satisfy audit requirements, demonstrate regulatory compliance, or provide a board-level assurance that AI agents cannot take unauthorized actions, deterministic enforcement carries significant procurement weight.

The SIEM and SOC integration capability reinforces the enterprise fit.

Feeding agent behavioral logs and anomaly detections into existing security tooling means that AI agent governance becomes part of the security operations workflow rather than a separate programme with separate tooling and separate reporting. That integration posture directly addresses the tool consolidation pressure that most enterprise security leaders are operating under.

For vendors adjacent to this space, including privileged access management platforms, identity governance providers, and AI infrastructure vendors, the Xage announcement is a competitive signal worth analyzing carefully. The buyer conversation is shifting from “how do we prevent AI from generating harmful content” to “how do we govern what AI agents can actually do inside our infrastructure.” Those are fundamentally different purchasing conversations with different ICP profiles and different budget origins.

The Governance Window Is Narrowing

Enterprise security program have a compressed window to establish AI agent governance frameworks before autonomous deployments reach a scale where retrofitting controls becomes genuinely difficult.

The current moment, where most organizations have AI agents in limited production or late-stage pilots, is precisely the right time to build governance architecture that scales forward rather than patches backward.

Identity assignment, policy scoping, behavioral baselining, and audit trail infrastructure are significantly easier to establish before an agent estate grows than after it has accumulated months of undifferentiated activity logs and broad, unreviewed access grants.

Xage’s platform addresses the architectural requirements of that governance window directly.

Whether it becomes the category-defining solution or one of several competitive approaches, the capability set it represents- deterministic agent visibility, identity-anchored access control, shadow AI detection, and SIEM-integrated anomaly surfacing- is precisely what enterprise security procurement conversations will increasingly demand as AI deployment accelerates through 2026 and beyond.

The agents are already inside the perimeter. The infrastructure to govern them is not yet standard. That gap is where the next significant wave of enterprise security investment is headed.

Research and Intelligence Sources: Xage Security

To participate in our interviews, please write to our CyberTech Media Room at info@intentamplify.com