ProvenantAI Get in touch
← Standards register Assessment

OWASP Top 10 for Agentic Applications

Risk taxonomy · 10 risks · 2026

A compromised server can't hand your agent a new tool. You chose the tools, and nothing else can add to them.

6

Enforced

Provenant does it. You can show it before your agent runs.

4

Contained

Where the standard asks you to prevent something the model might do, Provenant limits how far it gets.

RefRequirementHowMechanism
ASI01Agent Goal HijackContainedThe goal can be hijacked; what it can do can't
ASI02Tool Misuse and ExploitationEnforcedWhat your agent can do next depends on what it has done and who approved it
ASI03Identity and Privilege AbuseEnforcedYour agent can't grant itself anything
ASI04Agentic Supply Chain VulnerabilitiesEnforcedA compromised server can't hand your agent a new tool
ASI05Unexpected Code ExecutionContainedYou know everything that could run before it does
ASI06Memory and Context PoisoningContainedUntrusted content can only affect what you allowed it to
ASI07Insecure Inter-Agent CommunicationEnforcedA sub-agent can only send back what you allowed
ASI08Cascading FailuresContainedYou know how far a failure could spread before it does
ASI09Human-Agent Trust ExploitationEnforcedYou choose what the approver sees before they decide
ASI10Rogue AgentsEnforcedAn agent can't operate outside what you allowed