P.K. SHARMA

Cyber security intelligence, AI governance, practitioner analysis

Cloud and subscribed services

AI model endpoint (LLM API): what it is and how it is attacked

Where an application sends a question, along with documents, images or data, to an AI model and gets an answer back.

Also known as

  • LLM API
  • inference endpoint
  • model service
  • AI service

Typing any of them into the editor finds this object.

Why it matters on a security diagram

It handles untrusted input by design, and whatever consumes its answer usually trusts it. Where the model has been given tools or credentials of its own, its permissions are the blast radius.

How it gets attacked, and what reduces it

How it gets attacked

  • Instructions hidden in content the model later reads
  • Model output used to trigger actions without a check

What reduces it

  • Treat model output as untrusted input to whatever consumes it
  • Never let an answer trigger an action without a check in between
  • Give the model narrower access than the person asking, not wider

Where it sits

Group
Cloud and subscribed services · Things that only exist because somebody else runs the platform underneath them.
Whose side, by default
Ours · Belongs to the organisation the diagram is about.
Catalogue identifier
model-endpoint

Reviewed . CC BY 4.0.

Go deeper

Prompt injection pattern library · Free tool

Others in cloud and subscribed services

The cloud and subscribed services group lists all 12 of them side by side.

AI model endpoint on your own diagram

Open the editor, press N, and type LLM API. The object is placed and connected to whatever was selected, and Tab adds the next one already joined to it. Nothing is uploaded: the page is served with a Content Security Policy that forbids the browser from making any outbound request at all.

Open the diagram maker