Aviatrix Launches Clearance to Contain Compromised Cloud and AI Workloads

Aviatrix has launched Clearance, a cloud security offering designed to limit the damage a compromised workload can cause by restricting the destinations and other workloads it can reach. Its central proposition is operational as much as technical: security teams can test a restrictive policy against real traffic before allowing it to block production connections.

Announced October 26, Clearance combines outbound and lateral containment with an automated policy layer called Aviatrix Active Defense. The company is also making outbound containment, lateral containment, and intrusion prevention available to enterprises without a workload cap through a temporary response program running from November 1, 2026, to May 1, 2027.

For organizations deploying AI agents alongside conventional cloud applications, the underlying problem is increasingly consequential. An exploited service or manipulated agent may already have useful credentials and network access. Stopping the initial compromise remains essential, but the connections available afterward help determine whether that compromise stays local or spreads.

Default Deny With a Preview of the Consequences

Clearance runs on the Aviatrix Containment Platform, built on the company’s Cloud Native Security Fabric. It gives workloads an explicit set of permitted communication paths, with other paths denied by default.

Outbound Containment governs connections to external destinations, expressed by domain rather than only by network address. Lateral Containment governs communication between workloads by identity, extending the model across clouds and to resources such as containers, serverless functions, databases, and AI agents.

The deployment challenge is that legitimate application dependencies are often difficult to enumerate completely. A rule that appears correct during review can still interrupt a payment service, a background job, or an infrequently used integration.

Clearance addresses that risk with a Shadow Rule Set. A proposed policy first runs in monitor mode alongside the live policy, showing the traffic it would have denied without blocking those connections. Teams can identify required paths, revise permissions, and then switch to enforcement. Aviatrix says enforcement and rollback each require a single action.

This makes the preview an important part of the workflow, rather than a guarantee that a change cannot cause an outage. Traffic observed during one period may not include month-end processing, disaster recovery, or another rare but legitimate activity. Teams still need to choose observation periods and exceptions that reflect how their applications actually operate.

How Network Enforcement Fits Cloud Applications

On its product website, Aviatrix describes transparent insertion at the VPC level, preserving original addresses without requiring a host agent, SDK, or application code change. It also says policies can be represented through Terraform and Kubernetes custom resources, and orchestrated into security groups and Kubernetes NetworkPolicy where a gateway is unnecessary.

That matters because cloud infrastructure changes more quickly than a static address list. A container may move, a function may scale, and an application may span several networks. Expressing intent around the workload and its required destinations can make a policy more durable than treating every address change as a separate security decision.

The practical goal is to close communication paths an attacker could use without forcing every application team to adopt a new integration. Network containment nevertheless governs connectivity: an allowed connection can still carry harmful activity. It complements patching, identity controls, application security, and detection rather than making those controls unnecessary.

Active Defense Adds Automated Policy Maintenance

Aviatrix Active Defense is the offering’s autonomous layer. According to the announcement, its agents build a model of the customer’s environment, draft the policies applications need, evaluate them in monitor mode, and keep them current as workloads change. The resulting policies remain readable and editable by the security team.

“You cannot patch or detect your way out of this in the time you have, and this patch cycle runs into the holiday change freeze,” Aviatrix CEO Doug Merritt said in the announcement, explaining the company’s decision to offer a temporary containment backstop while customers assess and patch systems.

The website adds an important availability distinction: Active Defense is in free early access for design partners on Aviatrix Enterprise and is planned for Aviatrix Cloud. That should be distinguished from the availability of Clearance’s containment controls. The page also says the automation operates within team-defined guardrails and does not modify rules written by people.

The potential benefit is less manual work translating application behavior into enforceable policy. The corresponding operational requirement is to keep ownership clear: automatically drafted rules need a defined review and promotion process, especially when an application’s observed behavior may already include unwanted connections.

Why AI Agents Make Reachability More Important

Aviatrix’s broader AI workload security technology uses network flow data, DNS logs, and cloud asset inventory to discover agents, model endpoints, and related infrastructure. The company describes this as a way for security teams to identify AI use without depending entirely on developers instrumenting applications.

Discovery and containment solve different problems. Discovery helps establish which systems exist and where they connect. Containment determines which of those connections remain possible once a policy is enforced.

Consider an agent that needs a model provider and one approved internal service. Restricting its network paths can reduce the destinations available for data theft or further access if that agent is compromised. It does not establish that the agent’s prompts, reasoning, or authorized actions are safe; those require controls at other layers.

This distinction keeps the product’s AI relevance concrete. Clearance is about bounding what an agent can reach through the network, while Active Defense uses automation to help maintain those boundaries.

The Free Program and Standard Availability

Aviatrix says all enterprises qualify for its response program, covering outbound containment, lateral containment, and intrusion prevention across their estate without a workload cap from November 1 through May 1. AWS Marketplace enrollment opens November 1, 2026.

Teams can begin with a free Containment Assessment. The website distinguishes an initial browser-based self-score, which provides provisional results, from a subsequent architecture assessment using customer data. The initial score should therefore be understood as a starting point for investigation rather than a measured inventory of every reachable resource.

Under its standard offering, Clearance includes an assessment and 15 workloads in monitor or enforce mode without a time limit. Additional capacity is priced per workload, with pricing available from Aviatrix. The announcement says the two containment controls are delivered through Aviatrix Cloud and are also available on Aviatrix Enterprise.

The launch targets a persistent obstacle in enterprise security: organizations often agree that workloads should have less access, but hesitate to enforce restrictions when the consequences are uncertain. Clearance’s significance lies in connecting that security objective to a staged operational process—observe the proposed policy, resolve dependencies, enforce it, and retain a clear path back.