LLM Security Checklist
Prompt injection, data leakage, and the 8 checks every team misses when deploying large language models. One page, zero cost.
What you get
- 8 security checks most teams skip when shipping LLM-powered features to production
- Covers prompt injection, training-data leakage, model extraction, and output sanitization
- A pass/fail column so you can score your deployment posture in 10 minutes
- Mapped to the OWASP Top 10 for LLM Applications so you can cite the standard
LLM Security Checklist
Score your LLM deployment security. Check each item you pass, then download your completed checklist as a PDF.
| OWASP | Security check | What to look for | Pass |
|---|---|---|---|
| LLM01 | Prompt Injection | Test both direct injection and indirect injection via RAG-retrieved documents. Insert a canary instruction in a source doc and check if the model follows it. | |
| LLM02 | Data Leakage | Probe whether the model reveals training data, PII, API keys, or internal system details. Test with extraction prompts and membership inference attacks. | |
| LLM03 | Training Data Poisoning | Verify provenance and integrity of fine-tuning data. Check for unauthorized modifications to training datasets, embeddings, or vector stores. | |
| LLM05 | Supply Chain | Audit model provenance — base models, adapters, tokenizers. Verify checksums. Check for typosquatted packages in ML dependencies. | |
| LLM06 | Excessive Agency | Review what tools, APIs, and permissions the model can invoke. Apply least-privilege. Gate destructive actions behind human approval. | |
| LLM07 | Output Sanitization | Check what happens between model output and the user's screen. Can the model produce script tags, iframes, or redirect URLs that execute? | |
| LLM08 | Model Extraction | Rate-limit API access. Monitor for systematic queries designed to replicate model behavior. Restrict output logprobs if not needed. | |
| LLM10 | Denial of Service | Test for resource exhaustion via long inputs, recursive tool calls, or context-window stuffing. Set max token limits and per-user rate limits. |
Need a full review? Grab the AI/LLM Security Review Toolkit for $149 or book an AI security assessment.