# LLM application security checklist

> For reference purposes only. Not legal advice. Draft for review by a qualified professional.

> How to use this template. A checklist organized around the OWASP Top 10 for LLM Applications, in original wording. Check the current OWASP list. Replace every [placeholder], delete guidance notes and have a qualified professional review the result. Template version 1.0, last reviewed 2026-10-03. Certainty: simplified.

## Prompt injection

Are inputs and retrieved content treated as untrusted? Are instructions separated from data?

[Enter your text here]

## Sensitive information disclosure

Can outputs reveal personal, confidential or system data?

[Enter your text here]

## Supply chain

Are models, plugins and datasets from trusted, checked sources?

[Enter your text here]

## Data and model poisoning

Is training and retrieval data protected from tampering?

[Enter your text here]

## Improper output handling

Is model output validated before it reaches other systems?

[Enter your text here]

## Excessive agency

Does the system have only the permissions and actions it needs?

[Enter your text here]

## System prompt leakage

Do prompts hold secrets that should live elsewhere?

[Enter your text here]

## Vector and embedding weaknesses

Are retrieval stores access-controlled and monitored?

[Enter your text here]

## Misinformation

Are users told outputs may be wrong, and are high-stakes outputs checked?

[Enter your text here]

## Unbounded consumption

Are rate, cost and size limits in place?

[Enter your text here]

## Certainty notes

- Simplified: a general starting point. Rules differ by region.

> For reference purposes only. This site offers general guidance for learning and planning; it is not legal advice and creates no lawyer-client relationship. Rules change often, so check the primary sources and qualified counsel before you act.