Sumsub adds access controls for AI agents

Sumsub has expanded its Know Your Agent framework with the KYA Module for MCP, giving businesses configurable controls over what AI agents can access and execute inside company systems. The module supports agents built internally or deployed using models such as Claude or ChatGPT, placing a filtering layer between agents and connected data sources.

The module screens requests and removes personally identifiable and other sensitive information before it reaches an agent. A configurable rule engine sets restrictions based on data sensitivity, action type, department and other parameters. An activity log provides an auditable record of agent requests and actions, while transaction monitoring lets businesses establish limits for sensitive files and financial operations.

Requests involving flagged files or actions above defined thresholds can trigger a real-time liveness check to confirm that an authorized human is behind them. Businesses can add questionnaires or address verification for more sensitive actions. Organization-wide controls initially support Amazon Bedrock AgentCore Gateway, with additional MCP gateways planned.

The launch builds on Sumsub’s earlier agent verification capability, which links automation to a verified human identity, and its MCP integration for direct platform operation. The expanded framework addresses the visibility and authorization gaps that can arise when agents receive broad access to company systems.