🧑‍💻

OWASP LLM Top 10 Assessment

🔒 In your browser

Self-assess your LLM app against the OWASP LLM Top 10 — get a risk score and mitigations.

How it works

Answer a few questions about your LLM app or agent. You'll get a risk profile mapped to the OWASP Top 10 for LLM Applications (2025), prioritized by exposure, with concrete mitigations for each gap. Nothing is uploaded — it's computed in your browser.

100/100
Readiness (0/24 answered)
0 exposed
LLM01Prompt InjectionNot answered

Crafted input overrides instructions or smuggles commands via retrieved content.

Does the agent process untrusted input (user text, web pages, documents, emails)?
Do you separate trusted instructions from untrusted content (delimiting / structured messages)?
Do you screen inputs for known injection patterns before they reach the model?
LLM02Sensitive Information DisclosureNot answered

The model leaks secrets, PII, or proprietary data in its output.

Can the agent access secrets, PII, or other users' data?
Do you redact/scrub secrets and PII from prompts and context?
Do you filter model output for sensitive data before returning it?
LLM03Supply ChainNot answered

Compromised models, datasets, plugins, or MCP servers introduce risk.

Do you use third-party models, plugins, or MCP servers?
Do you pin versions and vet third-party tools/MCP servers before use?
Do you monitor tool/MCP definitions for silent changes (rug-pulls)?
LLM04Data & Model PoisoningNot answered

Tampered training/fine-tuning/RAG data biases or backdoors the model.

Do you fine-tune or feed a RAG store with externally-sourced data?
Do you validate and track the provenance of training/RAG data?
LLM05Improper Output HandlingNot answered

Model output is trusted downstream (SQL, shell, HTML, tool args) without validation.

Is model output passed to another system (DB, shell, browser, API, code)?
Do you validate/encode model output before using it downstream?
LLM06Excessive AgencyNot answered

The agent has more tools, permissions, or autonomy than the task needs.

Can the agent take actions with side effects (write, delete, pay, email)?
Are tool permissions scoped to the minimum needed (least privilege)?
Is there human approval for high-impact or irreversible actions?
LLM07System Prompt LeakageNot answered

Secrets or security-critical logic live in the system prompt and can be extracted.

Does the system prompt contain secrets, keys, or security rules you rely on staying hidden?
Do you assume the system prompt could become public and enforce controls elsewhere?
LLM08Vector & Embedding WeaknessesNot answered

RAG vector stores leak data across tenants or are poisoned via embeddings.

Do you use a vector store / embeddings for RAG?
Do you enforce per-tenant/per-user access control on retrieved chunks?
LLM09MisinformationNot answered

Confident but wrong output (hallucination) is relied upon in decisions.

Do users act on the agent's answers in high-stakes contexts (legal, medical, financial, code)?
Do you ground answers in sources and show citations/uncertainty?
LLM10Unbounded ConsumptionNot answered

No limits on tokens/tool calls/cost enable denial-of-wallet and DoS.

Is the agent exposed to untrusted or high-volume traffic?
Do you enforce rate limits, token/cost caps, and loop/step limits?

This maps to the public OWASP Top 10 for LLM Applications and is a self-assessment starting point — not a certification or a substitute for a professional security review.

🎯You might also need LLM Red-Team Prompt Generator

About the OWASP LLM Top 10 Assessment

The OWASP Top 10 for LLM Applications is the industry-standard list of the biggest security risks in AI apps and agents — prompt injection, sensitive-information disclosure, excessive agency, and more. This free self-assessment asks a few plain questions about your app and turns them into a risk profile: which OWASP LLM risks you're exposed to, a readiness score, and specific mitigations for each gap.

It runs entirely in your browser — nothing about your system is uploaded.

How to use it

  • Answer yes / no / unsure to the questions under each of the ten risks.
  • Watch the readiness score and per-risk status (Exposed / Partial / OK) update live.
  • Read the mitigations shown for each exposed risk, and copy the full report.

The OWASP LLM Top 10 (2025)

  • LLM01 Prompt Injection · LLM02 Sensitive Information Disclosure · LLM03 Supply Chain
  • LLM04 Data & Model Poisoning · LLM05 Improper Output Handling · LLM06 Excessive Agency
  • LLM07 System Prompt Leakage · LLM08 Vector & Embedding Weaknesses
  • LLM09 Misinformation · LLM10 Unbounded Consumption

Frequently asked questions

What is the OWASP Top 10 for LLM Applications?

It's a community-driven list, published by the OWASP GenAI Security Project, of the ten most critical security risks specific to applications built on large language models.

Is this a certification?

No. It's a self-assessment to help you find gaps and prioritise fixes — not a formal audit or certification, and not a substitute for a professional security review.

How is the readiness score calculated?

Each risk is scored by how many of its questions you answered in the risky direction (with 'unsure' counted as half). The score is the inverse of your average exposure across the risks you answered.

What do I do with an 'Exposed' risk?

Each exposed or partial risk lists concrete mitigations. Several link to ToolsHub tools — the MCP Security Scanner, Prompt Secret Redactor, Prompt Injection Checker and the Red-Team Prompt Generator.

Is anything uploaded?

No. Your answers and the report never leave your browser.

Related searches

owasp llm top 10llm security assessmentai security auditllm threat modelagent security checklistgenai security

Related Developer tools

🔗 Embed this tool on your website — free

Copy this and paste it into your page's HTML. The tool runs in the visitor's browser, just like here. Change height to fit, or add the optional auto-resize snippet below. Add ?theme=dark to the URL for dark mode.

<iframe src="https://toolhq.dev/embed/owasp-llm-top10-assessment/" title="OWASP LLM Top 10 Assessment — ToolsHub" width="100%" height="520" style="border:1px solid #e5e7eb;border-radius:12px;max-width:680px" loading="lazy"></iframe>
Optional: auto-resize the iframe height

Add this once on the same page so the iframe grows to fit the tool:

<script>
addEventListener("message", function (e) {
  if (e.data && e.data.type === "toolshub:resize") {
    document.querySelectorAll('iframe[src*="/embed/"]').forEach(function (f) {
      if (f.contentWindow === e.source) f.style.height = e.data.height + "px";
    });
  }
});
</script>