What is data exfiltration?
Data exfiltration is the unauthorized movement of data from inside an organization to a destination outside its control. Classic routes include files copied to personal cloud storage, records emailed to private addresses, and malware that sends data to an attacker. Through prompt injection, an AI agent that reads untrusted content can also be told to send data out through any tool that reaches the outside world, such as outbound email or a web request with data hidden in the address.
The risk is highest when one agent combines access to private data, exposure to content an outsider can influence, and a way to send information out. Removing any one of the three, such as the ability to email unapproved domains, closes that route.
Data exfiltration in practice
A precision parts manufacturer has an agent answer supplier emails using its drawings and quoting system. Hidden text in one message tells the agent to send a customer’s full drawing set to an outside address. The platform blocks mail to domains outside the approved supplier list, so the send fails and the audit log records the attempt.
What to watch for
- Block automatic loading of external images in AI chat interfaces, since a model can hide data in an image’s web address.
- Require human approval for outbound messages and uploads that carry internal files.
- Alert on sudden jumps in how much data an agent reads or sends.
- Do not assume that running the model privately stops exfiltration, because an agent’s outbound tools remain a route.