LLM06: Excessive Agency

Excessive Agency sits in the agent permissions surface, and OWASP LLM Top 10 rates it critical. Tools and plugins granted to an agent exceed the principle of least privilege. For teams shipping LLM and agentic features, a control like this is only as good as the evidence that it was actually tested - an unverified control is a finding waiting for an auditor.

Penaxtra ships adversarial probe families that target excessive agency directly. Each probe runs against your live endpoint on a schedule, and every triggering response is scored by three independent judges plus a meta-judge before it becomes a finding - so a LLM06 result is a tested verdict, not a guess. Every relevant finding is created with the OWASP LLM Top 10 LLM06 identifier already attached, so it lands in the audit-evidence pack mapped to the control rather than as a screenshot someone has to translate later. Where the same weakness touches another framework, the cross-framework overlap means one finding satisfies several control cells at once.

Findings for LLM06 carry the OWASP LLM Top 10 LLM06 identifier and cross-map to the related controls in the other five frameworks Penaxtra covers.