OWASP LLM Top 10 2026: The Full List and What Changed

OWASP’s 2026 LLM Top 10, released August 4 by its GenAI Security Project, ranks the most significant security risks for applications that use large language models as components and incorporates real-world incident data for the first time, weighted at 25% alongside 75% practitioner consensus. Prompt Injection and Sensitive Information Disclosure remain the top two risks, while Excessive Agency rose from sixth to third as systems increasingly receive permissions to perform consequential actions, and Unbounded Consumption and Misinformation also moved higher following incident evidence. The list also renamed System Prompt Leakage as Hidden Context Exposure to encompass all non-user-facing context, including retrieved policies and tool schemas, while Supply Chain Vulnerabilities, Data and Model Poisoning, Vector and Embedding Weaknesses, and Improper Output Handling declined in rank. OWASP distinguishes this framework from its separate Agentic Top 10, which addresses autonomous systems with tools, memory, and independent actions, though organizations may need both as LLM applications gain agency. Recommended practices include mapping data and output flows, testing for injection and data disclosure, restricting permissions, validating all model output before downstream use, protecting contextual information, automating security checks in delivery pipelines, and reassessing risks whenever models, tools, or data sources change.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.