Zero Trust for Agents: Implementing Context Lineage in the Enterprise Data Mesh

Challenge: When Agentic Bots Become Primary Data Reader

In large data platforms, AI agents now execute more data queries than human users. For teams that are running thousands of internal services, it is very common to have hundreds or thousands of agentic bots querying data: a “Supply Chain Optimizer” reading manufacturing logs, a “System Quality Analyst” agent querying usage metrics, or a “Sales Forecaster” aggregating regional sales data, finally passing or interacting with some models.

In a decentralized data mesh, domain owners need a way to detect whether an agent that they allowed to read critical data has been altered or compromised since its identity was issued. In such cases, mTLS authenticates the caller service but provides no details about the agent’s prior actions or execution context, such as which model or service it is, or what actions it has performed with the data in the past.