AI Worker governance readiness

AI GSAQ™

AI Governance Self-Assessment Questionnaire

Review whether an AI Worker has the ownership, bounded authority, data and tool controls, Human Review, evidence, monitoring, and lifecycle controls needed for its intended use.

Start the assessment Review the assurance path

Assessment protocol

A structured path from scope to accountable review.

AI GSAQ follows a four-stage review format. Select the operating profile, assess the evidence for each control area, review gaps, and hand the result to an accountable person. The questionnaire does not create authority.

01Set scopeIdentify the AI Worker and intended operating boundary.
02Select profileChoose the highest applicable authority and consequence profile.
03Review controlsAnswer 12 evidence-oriented questions across four control areas.
04Prepare decisionRoute gaps and supporting evidence to accountable Human Review.

Public review boundary: This page is not a completed customer assessment record, certification, compliance conclusion, security authorization, audit opinion, production approval, Passport activation, Runtime Permit, or action authorization.

AI GSAQ / Public self-assessment

AI Worker governance control review

Version 1.0 · 12 questions · Browser-local · Human decision required

Section 01

Operating profile

Choose the profile with the highest intended authority or consequence. If two profiles apply, select the more restrictive one.

A — AssistiveDrafts, summarizes, or recommends; no direct external or system action. H — Human-MediatedPrepares work that an accountable person reviews before use or release. E — External InteractionCommunicates with customers, partners, the public, or other external parties. X — Execution-CapableCan request or perform a controlled enterprise-system action. O — OrchestratedCoordinates multiple AI Workers, tools, workflows, or delegated tasks. D — ComprehensiveHigh-risk, sensitive, business-critical, regulated, or broadly autonomous use.

Control area 01

Accountability, purpose, and inventory

Confirm who owns the AI Worker, what it is allowed to do, and how it is classified.

GSAQ-01

Is a named human owner accountable for the AI Worker's purpose, operation, escalation, pause, and retirement?

Expected verification: Passport owner, sponsor, operator, reviewer, and escalation role assignments.

GSAQ-02

Are the business purpose, intended outcome, users, trust boundary, permitted work, and prohibited work explicitly defined?

Expected verification: Current Passport purpose and scope, prohibited actions, decision consequence, and success criteria.

GSAQ-03

Is the AI Worker inventoried with a risk tier, lifecycle state, review status, and reassessment date?

Expected verification: Registry entry, risk rationale, current lifecycle state, review date, and renewal trigger.

Control area 02

Data, models, and tool boundaries

Check that the AI Worker's information, dependencies, and technical capabilities are explicitly bounded.

GSAQ-04

Are approved data classes, sources, retention, memory, retrieval, privacy, residency, and redaction boundaries documented?

Expected verification: Data classification, source approvals, memory/RAG constraints, retention, and sensitive-data handling evidence.

GSAQ-05

Are models, providers, dependencies, limitations, shared responsibilities, and change controls reviewed?

Expected verification: Vendor Model Passport, provider review, dependency inventory, limitations, and change monitoring.

GSAQ-06

Are tools, APIs, connectors, actions, resources, credentials, destinations, and prohibited actions explicitly bounded?

Expected verification: Tool/API/Connector Registry, least-privilege permissions, credential custody, destination allowlists, and denial tests.

Control area 03

Human oversight and controlled execution

Verify that consequential activity cannot bypass policy, review, or action-time authority.

GSAQ-07

Do Toll Gates and Human Review block consequential, sensitive, uncertain, or out-of-scope work before action?

Expected verification: Policy and Toll Gate decisions, Human Review Items, reviewer accountability, separation of duties, and fail-closed tests.

GSAQ-08

For execution-capable profiles, must each action have current scoped authority and route through controlled execution?

Expected verification: Short-lived Runtime Permit, Action Broker mediation, exact action/resource binding, deny-by-default behavior, and outcome evidence.

GSAQ-09

Has the AI Worker been evaluated for prompt injection, excessive agency, unsafe delegation, tool misuse, exfiltration, and output failure?

Expected verification: Risk-based evaluations, adversarial tests, structured-output validation, failure analysis, and remediation records.

Control area 04

Evidence, resilience, and lifecycle

Confirm that decisions can be reconstructed and the AI Worker can be contained, reassessed, or retired.

GSAQ-10

Can reviewers reconstruct material decisions, actions, denials, exceptions, evidence, and outcomes without exposing sensitive content?

Expected verification: Workflow Events, Evidence Records, Stamps, decision records, integrity references, retention, redaction, and Audit Export.

GSAQ-11

Are monitoring, alerting, escalation, incident response, pause, disable, rollback, recovery, and restart controls assigned and tested?

Expected verification: Monitoring ownership, trigger thresholds, response runbooks, kill switch, rollback path, recovery rehearsal, and restart approval.

GSAQ-12

Do material changes, incidents, drift, expired evidence, provider changes, and scope changes trigger reassessment or lifecycle action?

Expected verification: Renewal cadence, material-change rules, evidence expiration, suspension/revocation criteria, and retirement records.

Final reviewCheck completeness and prepare the browser-local result.

Unanswered questions or missing critical controls will fail closed.