Back to Risk Catalogue

OWASP Top 10 for Agentic Applications

The ten most critical security risks for agentic AI applications, published by the OWASP Gen AI Security Project — Agentic Security Initiative (2026 edition).

ASI01: Agent Goal Hijack

Risks

AIR-SEC-010Prompt Injection

ASI02: Tool Misuse and Exploitation

Risks

AIR-SEC-024Agent Action Authorization Bypass AIR-SEC-025Tool Chain Manipulation and Injection

ASI03: Identity and Privilege Abuse

Risks

AIR-SEC-024Agent Action Authorization Bypass AIR-SEC-029Agent-Mediated Credential Discovery and Harvesting

ASI04: Agentic Supply Chain Vulnerabilities

Risks

AIR-SEC-026MCP Server Supply Chain Compromise

ASI05: Unexpected Code Execution (RCE)

Risks

AIR-SEC-025Tool Chain Manipulation and Injection

ASI06: Memory & Context Poisoning

Risks

AIR-SEC-027Agent State Persistence Poisoning

ASI07: Insecure Inter-Agent Communication

Risks

AIR-OP-028Multi-Agent Trust Boundary Violations

ASI08: Cascading Failures

Risks

AIR-OP-028Multi-Agent Trust Boundary Violations

ASI09: Human-Agent Trust Exploitation

Risks

AIR-OP-017Lack of Explainability

ASI10: Rogue Agents

Risks

AIR-OP-014Inadequate System Alignment AIR-OP-028Multi-Agent Trust Boundary Violations