Agent governance posture — and the open risks behind it.

How this is computed

Posture and findings are computed from observed actions — ownership, authorization stance, data exposure, guardrail coverage — not self-reported by the agents.

64/ 100

Needs attentionposture index

Dragged down by an unowned root agent acting freely and a default-allow stance — both let actions through with no identity check.

Posture by category

agent-native dimensions

Identity & OwnershipAt risk45

1 agent unowned · 1 runs as root and is taking actions.

AuthorizationAt risk40

Default authorization stance is ALLOW — actions permitted unless explicitly denied.

Data ProtectionNeeds attention55

1 agent sent PII (EMAIL_ADDRESS, US_SSN) to an unapproved model.

Guardrail CoverageNeeds attention60

3 of 4 guardrails on — but 'Require owner before action' is OFF.

Governed CoverageHealthy80

4 of 5 agents governed (in the MCP registry with a data scope).

Open findings

5 open · sorted by severity

highsupport-copilot exposed PII (EMAIL_ADDRESS, US_SSN) to an unapproved modelqscan sensitiveDataType · class: unapprovedowner · SupportOpen

highmarketing-gpt is ungoverned, unowned, runs as root, and is taking actionsgoverned · owner · sourceSystemUser (is-root)unassignedOpen

highDefault authorization is ALLOW — every agent action permitted unless denieddefaultDomainActionowner · SecurityOpen

mediumGuardrail 'Require owner/identity before action' is OFF@qpoint guardrail packagesowner · PlatformOpen

mediumdata-pipeline uses deprecated-tls on an internal actionriskLabels · actionowner · DataMitigating

4/5

governed

unowned

ALLOW

auth default

PII exposure

guardrail gap

Fields: sourceExe ownership · sourceSystemUser (is-root) · defaultDomainAction · action · qscan sensitiveDataType · @qpoint guardrail packages · riskLabels