Slide 3 of 27
Part 1 · What Is It?Slide 3
Slide 3 · The Definition, Part 1
Here's OWASP's actual sentence. Let's take it apart.
LLMs, especially when embedded in applications, risk exposing sensitive data, proprietary algorithms, or confidential details through their output.
“especially when embedded in applications”

The risk isn't really about the raw model in isolation. It's about what you connect that model to — databases, internal documents, live API calls, retrieval systems.

“risk exposing … through their output”

The disclosure path is the response text itself. Not a database breach. Not a stolen credential. The answer the model gives you.

“sensitive data, proprietary algorithms, or confidential details”

Three different things can leak: personal data, the model or business's own secrets, and anything else marked confidential.

The Result, In OWASP's Words

“This can result in unauthorized data access, privacy violations, and intellectual property breaches.”

← BackNext → The Definition Part 2