← Concept Index

Agent & LLM security

Confidential-data leakage

Also called: data exfiltration, oversharing

DEFINITION

Sensitive information leaving where it should stay — pasted into a prompt that's logged or used for training, echoed to the wrong user, or exfiltrated by an injected instruction through a tool the agent can call.

WHY IT MATTERS

It's the risk that turns a productivity tool into a breach. The defences are practical: minimise what you paste, know the product's data policy, and scope what an agent's tools can reach.

COMMONLY CONFUSED WITH

A hack of the model. Most leakage is mundane — someone pasted a customer list, or a tool had broader access than the task needed.

SOURCES

First-pass citations, limited to primary sources; a reviewer will broaden and verify these before this entry leaves draft.