Model output is untrusted: XSS and injection via rendering
Comes up in: Web fundamentals, AI system design, Frontend system design
This concept is being written. It gets the full five steps — explanation, predict, interview angle, practice with tests, and review cards. Here is exactly what it will cover, so you can study it elsewhere in the meantime.
What it covers
- OWASP Improper Output Handling (LLM10 in the 2026 edition, LLM05 in 2025 — number via
<Fact>) - markdown → HTML with raw HTML disabled
- sanitising (DOMPurify)
- link safety (
javascript:URLs, lookalike domains,rel) - images as exfiltration channels (markdown image URLs carrying data)
- CSP for AI pages
Words you will know
What you will practise
- break and then secure a naive markdown renderer
Interview questions this answers
- Why is rendering LLM output with
innerHTMLdangerous? - How can a markdown image in model output leak user data?