OWASP Agentic Top 10
The OWASP Agentic Top 10 is the security community's reference for the most common risks in agentic AI systems. Syntrix maps every check we run to one or more of these categories.
Prompt Injection
Untrusted input changes the agent's instructions.
Syntrix coverage: TOOL-01, INJ-01
Sensitive Information Disclosure
Agent leaks secrets, internal state, or system data.
Syntrix coverage: TLS-01, ERR-01
Supply Chain Risks
Malicious dependencies, models, or tool packages enter the agent.
Syntrix coverage: Not yet covered
Data and Model Poisoning
Adversarial training or runtime data warps agent behavior.
Syntrix coverage: Not yet covered
Improper Output Handling
Downstream systems blindly execute agent output.
Syntrix coverage: Not yet covered
Excessive Agency
Agent has more permissions, tools, or autonomy than the task needs.
Syntrix coverage: NET-01, AUTH-01, PERM-01, CORS-01
System Prompt Leakage
System prompts and operator instructions exposed to attackers.
Syntrix coverage: Not yet covered
Vector and Embedding Weaknesses
Embedding stores poisoned, leaked, or manipulated.
Syntrix coverage: Not yet covered
Misinformation
Agent confidently generates false claims that drive action.
Syntrix coverage: Not yet covered
Unbounded Consumption
Resource, cost, or rate limits absent on agent surface.
Syntrix coverage: SAMP-01, RATE-01
This list summarizes the OWASP Agentic AI Security Project. For the canonical version and updates, refer to OWASP directly.