ARAgentRiskLayerSecurity workspace
Assess

Assess an AI agent

Describe what is true in the current version. Planned controls do not count as implemented, and “I’m not sure” remains an information gap rather than a vulnerability.

About 5–10 minutesPrivate by defaultEvidence stays tied to one exact version
Agent & accessAbout the agent
Agent scopeAbout the agent

Which agent are you checking?

Assess one agent or one stable deployment at a time. Agents with different permissions should be checked separately.

Recommended for every assessment and required for “Other”. Do not paste secrets, credentials or customer data.
Assessment target

Freeze the exact repository version

Optional for a first questionnaire. Add both fields when this assessment should be tied to source evidence and exact retests.

When both are supplied, AgentRiskLayer records this exact target in the assessment evidence scope. A later commit must be assessed as a new revision.

Agent & access0%

Your answers are security-sensitive. They are submitted over the current site connection and assessments remain private unless you explicitly enable sharing. If you are not signed in, keep the private result link safe and sign in from the result page if you want to claim the assessment to your account.