Data Provenance Tracking

A technical concept in AI agent security referring to the capability of tracing the origin of tool call parameters. By maintaining taint labels on parameters, it distinguishes among direct user input, system prompts, and external data, enabling assessment of the security risk level associated with a given tool invocation.