# AI Security Review vs LLM Quality Audit

**Answer first:** an [LLM Quality Audit](/llm-quality-audit) measures how well the system does its job. An [AI Security Review](/ai-security) asks whether someone can make it leak data, bypass access rules, or act beyond its permissions. They answer different questions; one does not replace the other.

| | AI Security Review | LLM Quality Audit |
| --- | --- | --- |
| Core question | Can it be abused? | Does it perform? |
| Typical checks | Prompt injection (including via retrieved docs), retrieval ACL / cross-tenant exposure, secrets in answers, tool misuse, unsafe output channels | Task success, groundedness, golden set, refusal both ways, cost and latency gates |
| Best when | The assistant reads internal docs, serves multiple roles/tenants, or can call tools | You need measured quality scores and a repeatable eval suite |
| Starting format | One application, agreed workflows, scoped after intake | Fixed-scope audit from €8,000 (€6,800 co-branded startup option) |
| Primary outputs | Decision brief, access map, findings with repro, coverage table, fix list, optional retest | Scores, failure modes, eng report, exec summary, suite you can keep |
| Standards language | OWASP LLM / Agentic Top 10 (mapping, not certification) | Internal rubrics and agreed pass marks |

## When to book which

- **Security first** if the near-term risk is leakage, poisoned knowledge, or an agent that can write or call APIs.
- **Quality first** if the near-term risk is wrong answers, weak grounding, or no CI gates.
- **Both** when a customer or board asks for production readiness and AI cybersecurity evidence on the same system — run as separate scopes so findings stay clear.

## Related

- [AI Security Review vs classic penetration test](/ai-security/vs-pentest)
- [OWASP LLM Top 10 2026 — what we check](/guides/owasp-llm-top-10-2026)
- [RAG track on AI Security Review](/ai-security#rag-track)
- [Readiness vs Audit](/guides/readiness-vs-audit) (quality ladder)

**Next:** [Request AI Security Review](/ai-security#scope) · [LLM Quality Audit](/llm-quality-audit#scope) · [All guides](/guides)
