Internal data protection systems and methods within the execution flow of AI agent platforms, including machine-readable data records.

Abstract

The invention relates to a system and method for protecting internal data within the execution flow of an AI agent platform's tool. When the system works with a public large language model (LLM), the system sends only minimal state information that reflects the state of the tool's execution process and does not contain any data values, records, fields, or content from the original execution results of the internal system, preventing the organization's internal raw data from leaking to public LLMs while ensuring that these models have sufficient information to make subsequent decisions.

Images