OWASP Top 10 for LLM Applications
The international standard for vulnerabilities affecting Large Language Model applications, autonomous agents, and RAG architectures.
What's New in the 2025 Top 10
Reflecting modern RAG architectures, autonomous agent deployments, and cloud financial exposure.
Unbounded Consumption
Expands beyond traditional DoS to address GPU resource depletion and catastrophic cloud bills from "Denial of Wallet" (DoW) attacks.
Vector & Embedding Weaknesses
Dedicated protection for Retrieval-Augmented Generation (RAG), vector databases, embedding inversion attacks, and cross-tenant context leaks.
System Prompt Leakage
Counters extraction of confidential system preambles, embedded credentials, role boundaries, and internal business logic.
Excessive Agency
Broadened for autonomous multi-agent systems, unvetted tool integrations, and permissions that enable destructive real-world actions.
The OWASP Top 10 (2025)
LLM Application Architecture & Threat Map
Map the OWASP Top 10 vulnerabilities directly onto your application components: from Ingress prompts to Model inference, Vector Databases, and Tool extensions.
LLM Security Controls Checklist
Audit your AI application against official OWASP mitigations. Progress is saved locally in your browser.