Back to Risk Catalogue

OWASP LLM Top 10

The ten most critical security risks for applications built on large language models, published by OWASP.

LLM01:2026 Prompt Injection

Risks

RI-10Prompt Injection RI-25Tool Chain Manipulation and Injection RI-29Agent-Mediated Credential Discovery and Harvesting

LLM02:2026 Sensitive Information Disclosure

Risks

RI-01Information Leaked To Hosted Model RI-02Information Leaked to Vector Store

LLM03:2026 Excessive Agency

Risks

RI-10Prompt Injection RI-18Model Overreach / Expanded Use RI-24Agent Action Authorization Bypass RI-29Agent-Mediated Credential Discovery and Harvesting

LLM04:2026 Supply Chain

Risks

RI-08Tampering With the Foundational Model RI-09Data Poisoning

LLM05:2026 Data and Model Poisoning

Risks

RI-10Prompt Injection RI-09Data Poisoning

LLM06:2026 Unbounded Consumption

Risks

RI-10Prompt Injection RI-07Availability of Foundational Model

LLM07:2026 Misinformation

Risks

RI-20Reputational Risk RI-04Hallucination and Inaccurate Outputs RI-05Foundation Model Versioning RI-06Non-Deterministic Behaviour

LLM08:2026 Hidden Context Exposure

Risks

RI-14Inadequate System Alignment

LLM09:2026 Vector and Embedding Weaknesses

Risks

RI-02Information Leaked to Vector Store

LLM10:2026 Improper Output Handling

Risks

RI-09Data Poisoning

LLM03:2025 Supply Chain

Risks

RI-30Skill/Plugin Supply Chain Compromise

LLM06:2025 Excessive Agency

Risks

RI-30Skill/Plugin Supply Chain Compromise