AIBound

Home / Glossary / OWASP Top 10 for LLM applications

Glossary

OWASP Top 10 for LLM applications

The OWASP Top 10 for LLM Applications is a community-written list of the ten most serious security risks in applications built on large language models. The current version is the 2025 list, which OWASP released on 17 November 2024. It starts with prompt injection and ends with unbounded consumption.

By the AIBound team · Updated 17 November 2026

Key takeaways

  • The current version is the OWASP Top 10 for LLM Applications 2025, released on 17 November 2024.
  • It is written for developers and security professionals who build and protect LLM applications.
  • OWASP publishes a separate Top 10 for agentic applications, so the LLM list is one of two to read.

Why the OWASP Top 10 for LLM applications matters

Security teams need a shared list of what can go wrong before they can test for it. OWASP's list gives them one. It began in 2023 as a community effort to name the security risks that are specific to AI applications, and OWASP has updated it since.

The list is widely used as a common reference. When a customer asks how an AI feature was tested, an answer that follows the ten items is one the customer can check against a public source.

How to use the OWASP Top 10 for LLM applications

The 2025 list names ten risks, in order: Prompt Injection, Sensitive Information Disclosure, Supply Chain, Data and Model Poisoning, Improper Output Handling, Excessive Agency, System Prompt Leakage, Vector and Embedding Weaknesses, Misinformation and Unbounded Consumption. A team can put it to work in four steps.

  1. 1 List the LLM applications the company builds and the ones its employees use.
  2. 2 Check each application against the ten items and note which ones apply.
  3. 3 Set a control and a test for each item that applies.
  4. 4 Repeat the check when the application changes or OWASP updates the list.

OWASP Top 10 for LLM applications vs the Agentic Top 10

OWASP publishes both lists. The LLM list covers applications built on large language models. The agentic list covers AI agents that plan and take actions.

OWASP Top 10 for LLM applications vs the Agentic Top 10
Question OWASP Top 10 for LLM applications OWASP Top 10 for Agentic Applications
What does it cover? Applications built on large language models Agentic AI applications that act with autonomy
Which version is current? The 2025 list, released 17 November 2024 The list for 2026, published 9 December 2025
What is the first item? LLM01:2025 Prompt Injection ASI01 Agent Goal Hijack

Sources

Sources checked 17 November 2026

Examples

OWASP Top 10 for LLM applications examples

FAQ

Frequently asked questions

What is the OWASP Top 10 for LLM applications?

It is a list of the ten most serious security risks in applications built on large language models. OWASP, an open security community, writes and updates it. The current version is the 2025 list, released on 17 November 2024, and it is aimed at developers and security professionals.

What are the OWASP Top 10 risks for LLM applications in 2025?

In order, they are Prompt Injection, Sensitive Information Disclosure, Supply Chain, Data and Model Poisoning, Improper Output Handling, Excessive Agency, System Prompt Leakage, Vector and Embedding Weaknesses, Misinformation and Unbounded Consumption. OWASP numbers them LLM01:2025 to LLM10:2025 and publishes a page of detail and mitigations for each one.

Is the OWASP Top 10 for LLM applications a compliance standard?

No. It is a community list that raises awareness and guides testing. No law requires it and nobody is certified against it. Customers and auditors still ask about it often, because it gives both sides the same ten names for the risks of an LLM application.

How is the LLM Top 10 different from the classic OWASP Top 10?

The classic OWASP Top 10 covers security risks in web applications in general. The LLM list covers risks that come from building on a large language model, such as prompt injection and excessive agency. A team that ships an AI feature inside a web application usually needs to read both lists.

Stop the risky AI your employees already run.

AIBound blocks the high-risk AI on every employee device, in real time. The 14-day trial needs no proof of concept.

Book a Demo