VaikoraVaikora

Vaikora › Blog › Product Updates

AI agents call tools and APIs without asking. Here is where Vaikora puts the gate.

Product Updates · Nov 14, 2026 · 4 min read

When an AI agent decides to call an external API, write a file, or query a database, nothing in the agent itself stops it from doing that. The agent reasons, produces an action, and executes. If the action is wrong, expensive, or in violation of policy, you find out after the fact.

That is the structural problem Vaikora solves. It sits inline between the agent and whatever the agent is about to call. Before the action executes, Vaikora evaluates it against policy and returns one of four decisions: allow, block, modify, or require approval. The agent either proceeds, is stopped, gets a modified version of its request, or waits for a human. No action reaches a tool or model without passing through that gate first.

Why inline placement changes the control model

Detection after execution gives you a log. Inline enforcement before execution gives you control.

Consider a concrete example. An agent is processing customer support tickets and has been given access to a CRM API. The agent decides to query a record that includes fields covered under GDPR. Without a runtime gate, that query runs. With Vaikora inline, the action is evaluated before it executes. If the agent's context does not satisfy the policy condition for that data class, the action is blocked. The agent receives a response explaining the block. The whole exchange takes milliseconds: Vaikora's enforcement decision runs at P50 8ms and P99 45ms.

The policy decision is deterministic. Identical input always produces identical output. There is no probabilistic scoring, no threshold to tune, no model that drifts. The rule evaluates and the decision fires.

What gets inspected and what gets written to the log

Vaikora ships six content inspection modules. These cover the content of a request, the tool being called, the identity of the agent, the data classification of involved fields, the action type, and contextual signals like session history. Policy rules can reference any combination of these.

Every decision, whether allow, block, modify, or require approval, is written into a SHA-256 hash-chained audit log. Each entry is signed into the chain, so the log is tamper-evident. If a compliance auditor asks what your agents did with PCI-scoped data last quarter, you pull the log. The chain proves nothing was altered after the fact.

Vaikora ships compliance presets for HIPAA, PCI-DSS, GDPR, SOC 2, and ISO 27001. A team trying to get AI agents into a PCI-scoped environment does not write those rules from scratch. They load the preset and adjust for their specific tool surface.

How Vaikora connects to the rest of your security stack

Enforcement events do not stay inside Vaikora. High-severity AI agent risk gets pushed into your existing tools automatically.

For teams running Microsoft 365, Vaikora's email security component connects through Microsoft Graph, runs analysis in the customer's own Azure subscription, and reports into the customer's own Sentinel workspace. No MX record change is needed.

Vaikora adds a runtime gate on top of Azure RBAC, AWS IAM, and Snowflake roles. It does not replace those controls. It enforces AI-specific policy above the layer those platforms operate at, because RBAC was designed for human users making deliberate requests, not agents issuing hundreds of tool calls per session.

What to check in your current setup

If your team is running AI agents against production systems, start with these questions.

Data443's Vaikora is available on AWS Marketplace and in the Azure Sentinel content hub. The open-source gateway, `vaikora-guard-mcp`, is published on PyPI under the MIT license and can be self-hosted. Start by mapping the tool surface your agents currently have access to, then identify which of those tools touch data classes that carry regulatory or business risk. That scope is where inline enforcement pays off first.

See Vaikora enforce policy on your AI

Open-core AI runtime control. Self-host the MIT gateway free, or run the hosted Control Plane.

Get a demo Self-host the gateway

More from the Vaikora blog