When AI agents pass data along, no one can tell where it came from

A new article on MIT Technology Review makes a simple warning: when one AI agent passes data to another, a company may not be able to tell where that data came from. The piece was produced by Insights, the magazine's custom content arm, not its editorial staff.

The example

A sales rep asks an AI agent for customer pricing. The agent pulls from restricted pricing systems, internal communications and archived quotes. A second agent then uses that data to draft a proposal. No person checks either handoff.

Later, the proposal turns out to contain unapproved, confidential and sensitive information. No one can trace where the bad data entered or how it spread across systems, files and agents.

The argument

The article says the problem is not any one tool. It is a visibility problem. Data is constantly copied, summarized, changed and reused in new forms. Security built only around apps, networks or devices is no longer enough, it says.

Its answer is data lineage. That means tracking where data came from and how it was copied, changed and reused. The article calls lineage the new security perimeter, meaning the main line a company has to defend.

What is not known

The example is an illustration, not a reported incident. The material does not name specific products, companies, costs or steps for setting up lineage tracking.