Aviatrix launches a containment platform for AI agent security

Aviatrix has launched a new Containment Platform for AI agents, with Zero Trust for AI Workloads now generally available and Aviatrix AgentGuard in early access. The original source is Aviatrix's press release, published through GlobeNewswire on April 29, 2026 at 06:00 ET.

This is a CISO story, not just a networking story. Aviatrix's core argument is that an AI agent is effectively a machine identity that can make autonomous decisions about which systems, APIs and data sources to reach next. If that agent is compromised, misconfigured or manipulated through prompt injection, the important question becomes: how far can it actually move?

Zero Trust for AI Workloads is meant to give enterprises network-level control over which external AI services and internal resources a workload can communicate with. Aviatrix describes three main controls: allow and deny lists for external AI services, default-deny controls against shadow AI, and policy enforcement across regions and workloads. The company also says it is publishing validated containment architectures for AWS Bedrock Agents, Azure AI Foundry Agents and enterprise MCP.

AgentGuard, now in early access, is designed to discover AI agents across VMs, Kubernetes and serverless environments, map the models, tools and data they connect to, and build a continuous risk profile. That matters because many companies are now building agents faster than their security architecture can keep up. Agents often receive access to the same internal systems as employees, but without the same organizational friction.

For enterprise leaders, the takeaway is blunt: agentic AI turns blast radius into a board-level issue. Logging, DLP and model policies are not enough if a compromised agent can still reach CRM, ERP, file systems, developer tools and external services from the same workload. Network controls, identity and AI governance need to be designed together.

The next step for CIOs and CISOs is to build an agent register that is tied to real infrastructure controls: which agents exist, which MCP servers or tools they can call, which data they can read and write, which external endpoints they can reach, and how they are shut down. Aviatrix's launch does not prove that one vendor has the answer. It does show where enterprise AI security is moving: from detecting everything after the fact, to limiting the blast radius before damage spreads.

Source: Aviatrix via GlobeNewswire, April 29, 2026: "Aviatrix Operationalizes the Containment Era with the Industry's First Containment Platform for AI Agents".