Skip to content

OWASP LLM Top 10

The OWASP LLM Top 10 is a community-maintained list of the ten most critical security risks in applications built on large language models. Published by the Gen AI Security Project at the Open Worldwide Application Security Project (OWASP), it covers LLM software the way the original OWASP Top 10 covers web applications.

The current 2026 revision, formally titled the OWASP Top 10 for LLM Applications 2026, names these risks in rank order:

  • Prompt injection
  • Sensitive information disclosure
  • Excessive agency
  • Supply chain
  • Data and model poisoning
  • Unbounded consumption
  • Misinformation
  • Hidden context exposure
  • Vector and embedding weaknesses
  • Improper output handling

Each entry documents the risk, its attack scenarios, and mitigations such as guardrails, least-privilege tool access, and output validation. Security teams use the list as a coverage checklist for threat modeling and red teaming.

First released in 2023, the list is revised as new risks surface. The 2025 edition dropped model theft and added system prompt leakage, along with vector and embedding weaknesses, a category covering RAG pipelines and vector database storage.

The 2026 edition kept all ten categories but reordered eight of them, renaming system prompt leakage to hidden context exposure so it covers any non-user-facing context an attacker can read back. Excessive agency climbed to third as applications handed models more tools to act with, while improper output handling fell to tenth.

A separate OWASP Top 10 for Agentic Applications, released in December 2025, covers autonomous agent systems under its own ASI01 through ASI10 numbering, addressing risks like goal hijacking and rogue agents that have no single-turn equivalent.

A woman holds a large padlock on a crate inside a shield dome while UFO beams bounce off it, and a conveyor feeds crates into a defense console.

Tutorial

Supply Chain Security for the Age of AI-Speed Exploits

Learn Python supply chain security as AI speeds up attacks: what the LiteLLM backdoor teaches, how to check your setup, and habits that last.

intermediate best-practices tools

For additional information on related topics, take a look at the following resources:

Have a question about this? Mentor AI can show you examples, compare related terms, and point you to tutorials.


By Martin Breuss • Updated Nov. 15, 2026