Improper AI Output Handling - OWASP LLM05

So you've decided to use an LLM to generate HTML and render it on a web page. Before you get too far, we should talk about sanitizing your outputs. LLM-generated output is untrusted input. If you put it into the DOM, into a SQL query, or into any execution context, you must treat it the same way you would treat anything coming from a user. We're all familiar with Little Bobby Tables. The same rules still apply. The Vulnerability LLMs can be manipulated through prompt injection to...