Enterprises Need Data Lineage for Agentic AI Security as 2 Unchecked Agents Spread Sensitive Data
Updated
Updated · MIT Technology Review · Nov 4
Enterprises Need Data Lineage for Agentic AI Security as 2 Unchecked Agents Spread Sensitive Data
3 articles · Updated · MIT Technology Review · Nov 4
Summary
Enterprises using agentic AI need data lineage to secure workflows, after a sample sales process showed confidential pricing data entering an AI-generated proposal with no way to trace its origin.
Two AI agents handled the task without human review: one pulled information from restricted systems, internal communications and archived quotes, and another used that output to draft the proposal.
That failure reflects a broader shift in enterprise risk, as AI systems continuously ingest, copy, summarize, transform and redistribute fragmented data across human, agentic and hybrid workflows.
Traditional security built around applications, networks, endpoints or static files is no longer sufficient; governance at scale now depends on visibility into how data moves, spreads and is reused.