OWASP LLM Top 10
The OWASP Top 10 for LLM Applications is a community-driven list of the most critical security risks specific to software built on large language models (LLMs) — the AI systems behind chatbots, copilots, and generative features. Maintained by OWASP's GenAI Security Project, it names the failure modes that show up again and again when you put an LLM into a real product. If you're shipping an LLM or generative-AI feature and want to know what to defend against, here's the plain-language version: what it is, who it's for, all 10 risks in the 2025 list, how it fits with governance frameworks, and how to get ready.
OWASP LLM mitigation checklist
A risk-by-risk hardening checklist.
Free readiness assessment
See your LLM security gaps in minutes.
Map to AI governance
How the Top 10 lines up with NIST AI RMF & ISO 42001.
Illustrative statuses. Your real assessment is generated from your answers in the app. Standpoint is a self-assessment aid, not legal advice.
What is the OWASP Top 10 for LLM Applications?
It's a prioritized risk list, not a certification. OWASP (the Open Worldwide Application Security Project) is best known for its long-running web-application Top 10; as teams began embedding large language models into products, the same community built a companion list for the risks that are specific to LLMs. The work lives in OWASP's GenAI Security Project and is refreshed as the technology and its threats evolve — the current edition is the 2025 list. Each entry describes a class of risk, why it matters, and the kinds of defenses that address it, so engineering and security teams can reason …
Who is it for?
Anyone building or securing software with an LLM inside it. In practice that means: - Developers integrating an LLM, a retrieval pipeline, or agent/tool-calling into an application; - Security teams and AppSec engineers who need to threat-model and test generative-AI features; - Product and platform teams shipping copilots, assistants, or any user-facing generative capability; - Governance, risk, and compliance (GRC) owners mapping AI risk to their broader program.
The 2025 OWASP Top 10 for LLM Applications
Here are the ten risks in the 2025 edition, each summarized in our own words. The one-line descriptions are Standpoint's paraphrase — see the official OWASP materials for the authoritative text. The above is Standpoint's own summary of the 2025 list. For authoritative descriptions and mitigations, consult the official OWASP Top 10 for LLM Applications from the OWASP Foundation.
How it complements NIST AI RMF and ISO 42001
These frameworks answer different questions and work well together. The OWASP Top 10 for LLMs is an application-security lens — concrete, technical failure modes to defend against in the product itself. NIST AI RMF (the U.S. AI Risk Management Framework) and ISO/IEC 42001 (the AI management-system standard) are a governance lens — how an organization governs, maps, measures, and manages AI risk across its lifecycle. In practice you use the governance frameworks to run the program and the OWASP list to make sure the engineering-level risks are actually being tested and mitigated. See the OWASP …
A practical path to readiness
- Map your LLM features. List where models, prompts, retrieval, and tool/agent access live in your product. - Threat-model against the Top 10. For each feature, ask which of the ten risks apply and how badly. - Run a gap assessment. Compare your current defenses to the mitigations for each relevant risk. - Harden and add guardrails. Put input/output handling, least-privilege scoping, rate limits, and human oversight in place. - Test and monitor. Red-team the features, log model behavior, and keep the controls current as the app evolves. The fastest start is a gap assessment. Our free OWASP LLM Top 10 readiness assessment shows where you stand, and the mitigation checklist turns it into a working plan.
Frequently asked questions
Q: Is the OWASP Top 10 for LLMs a standard? A: No. It's a community-maintained awareness and engineering guide from the OWASP Foundation's GenAI Security Project — a prioritized list of risks, not a formal standard or something you get certified against. It's meant to inform how you build and test, and to complement governance frameworks like NIST AI RMF and ISO 42001. Q: Who should use it? A: Developers, security and AppSec teams, and product teams shipping any LLM or generative-AI feature. GRC and risk owners also use it to connect application-level AI risk to their wider program. Q: How i…
OWASP LLM Top 10 work counts elsewhere too
The crosswalk maps your OWASP LLM Top 10 evidence onto the frameworks it overlaps — so you move forward on several at once.