Data Overexposure
LLMs reveal internal or sensitive data in responses when tool boundaries or filters fail.
Definition
Data Overexposure is an attack on the output layer of the Model Context Protocol (MCP). It occurs when an agent includes sensitive information—such as internal identifiers, user data, or infrastructure details—in its final response. This can happen due to overly broad tool outputs, missing filtering logic, or agents blindly including context retrieved from tools or memory. The result is leakage of data that should have remained private.
This attack lives in the output layer of the MCP model, where context is surfaced to the user without proper sanitization or authorization checks.
How MCP Security Helps
Akto prevents data overexposure by inspecting agent responses for sensitive values such as PII, secrets, or internal metadata. It applies redaction rules to outgoing content, tests filtering logic on tool outputs, and flags any responses that include high-risk fields without proper access validation.
