As AI agents connect to file storage, communication platforms, and financial systems, they routinely request broad access to complete tasks, often with limited visibility for the businesses deploying them. Sumsub, a UK-based provider of verification and compliance infrastructure, has expanded its Know Your Agent (KYA) framework with the launch of the KYA Module for MCP. The module lets businesses define what the AI agents they build or deploy, whether in-house or on models such as Claude or ChatGPT, can access inside company systems.
Agent security incidents put access controls in focus
According to research from the Cloud Security Alliance, 65% of organisations experienced at least one security incident caused by an agent operating on their network in the past year, with sensitive data exposure the main cause. In July 2026, Hugging Face reported that an OpenAI model acting as an autonomous agent breached its production systems and accessed internal datasets without a human authorising the action.
The new module builds on two earlier steps in the KYA framework. In January 2026, Sumsub launched AI Agent Verification, binding AI-driven automation to a verified human identity. In June 2026, it introduced a Model Context Protocol (MCP) integration that allows agents to configure and operate the platform directly. Mick Amelishko, AI Advocate at Sumsub, said the convenience of AI agents should not come at the cost of control over company data.
Filtering, monitoring, and liveness checks
The module sits between an agent and the data it connects to. It screens every agent request, removes personally identifiable and other sensitive information before it reaches the agent, and applies a configurable rule engine. Businesses can trigger defined actions based on data sensitivity, the type of action, the requesting and calling department, and other parameters. An agent activity log provides an auditable record of what each agent has requested and done.
When an agent tries to access a flagged file or take an action above a set threshold, such as approving an invoice above a defined limit, the system can trigger a real-time liveness check to confirm that an authorised human is behind the request. Businesses can add further verification steps, including questionnaires or address verification, depending on the sensitivity of the action. Organisation-wide rules are supported starting with Amazon Bedrock AgentCore Gateway, with additional MCP gateways to follow.
