ProductOWASP Top 10 for Agentic Applications
Hold your agent up against the OWASP Agentic Top 10
OWASP lists 10 security risks for AI agents with tools. Enoki groups each run of your agent by them, so you see where an attack worked.

What ASI01 to ASI10 look like in an agent
- ASI01
Agent Goal Hijack
Hidden instructions in a message or document steer the agent toward an attacker's goal.
- ASI02
Tool Misuse and Exploitation
The agent uses a tool it already has in a way that causes harm.
- ASI03
Identity and Privilege Abuse
The agent acts beyond its user's access, often with borrowed or leaked keys.
- ASI04
Agentic Supply Chain Vulnerabilities
An MCP server, tool or prompt the agent loads while running was tampered with.
- ASI05
Unexpected Code Execution (RCE)
A plain-language request gets the agent to run code or commands it should not.
- ASI06
Memory & Context Poisoning
False notes planted in the agent's memory shape its later answers and actions.
- ASI07
Insecure Inter-Agent Communication
Forged or altered messages between agents send them in the wrong direction.
- ASI08
Cascading Failures
One agent's error spreads through the agents and systems connected to it.
- ASI09
Human-Agent Trust Exploitation
A confident, polished answer from the agent leads a person to approve something harmful.
- ASI10
Rogue Agents
An agent drifts from its purpose, acts on its own and hides what it does.
Questions
Does the free assessment group results by OWASP risk?
No. The free assessment shows each finding and the chat behind it. Grouping by OWASP risk and the OWASP Agentic report need a paid plan.
Is there an OWASP Agentic Top 10 certification?
No. OWASP publishes the list as guidance and certifies no agent or product. Enoki's report shows how often your agent is tested and what was found, and it says it is not a certification.
How does the Agentic Top 10 differ from the LLM Top 10?
OWASP wrote the Agentic list for agents that use tools and act on their own. The OWASP LLM Top 10 covers the model and the app built around it, starting with prompt injection.
Point Enoki at your agent
Paste the URL your users reach. Use a staging copy if the agent can move money or delete data.