Skip to content
ML Academy

Introduced on day 53

Data exfiltration (in LLM applications)

Data leaving the system through a channel the model can use, such as a tool argument, an email recipient, a link or an image the interface fetches automatically.

In detail

A model that has seen private data and can emit a URL or call a send tool can be talked into encoding that data into the output. Close channels in code: allow-list hosts, disable auto-loaded remote resources, authorise recipients, and control who can read traces and logs.