How Governed Is Your AI Agent Stack?
A self-assessment across 7 governance dimensions. Mark each checkbox for each statement that is true for your organization today. Every unchecked box is an open gap. Get your final score and find out how to close them with actionable next steps.
25
Technical checks
7
Governance dimensions
4
Score tiers
01 · Agent Discovery
Agent Inventory and Visibility
0/4
Can you enumerate every agent, every user and every connection in one place?
We maintain a central registry of every AI agent in use across the org (Cursor, Claude, ChatGPT, Copilot and others)
New agent adoption is detected automatically. We do not rely on employees self-reporting when creating or using an agent
For every agent, we know exactly which third-party apps and services it can access
Each agent session ties to the employee who delegated it. We can trace any action back to a named user
02 · Credential Mediation
Credential Isolation
0/4
Do your agents hold real tokens? Could an attacker extract them?
Agents never hold real credentials (OAuth tokens, API keys, PATs) directly. If an agent's session is compromised, no usable credential can be extracted
Credentials are managed separately from the agent runtime. The agent never handles or stores the real token
Every agent's access traces back to one named employee. No agent operates under a shared or anonymous identity
Agent access expires when the task ends. No credentials or permissions persist between sessions
03 · Policy Engine
Runtime Policy Enforcement
0/4
Does every tool call evaluate against a policy before it executes?
Every agent action evaluates against a written policy before it reaches the downstream app. No action runs unchecked
Agent access approval and disapproval are taken based on defined policies
Policies evaluate the full context (who is asking, which app, what action type) and produce a deterministic outcome. No probability scores, no ambiguity
When multiple rules apply to the same action, there is a clear hierarchy. Overlapping policies do not produce unpredictable results
04 · Suspend-and-Resume
Human-in-the-Loop Controls
0/3
Can a human stop a sensitive action before it executes?
When an agent attempts a sensitive action (creating records, sending messages, deleting data), it pauses and a human reviews before execution
Approval prompts appear where the user is already working. The reviewer does not need to switch to a separate app or dashboard to approve or deny
When an action is denied, it is fully blocked. The downstream app never receives the request
05 · Dynamic Agent Context
Privilege Scope and Session Management
0/3
Do you apply least privilege to your agents?
Agents start every task with zero access. They do not inherit standing permissions from a prior task or a shared service account
Agent permissions are bound to the delegating user's identity. An agent acting on behalf of a marketing manager cannot access engineering repos
Low-risk actions (reading data, searching) flow with minimal friction. High-risk actions (writing, deleting, sending externally) require additional approval or are blocked by default
06 · HRIS/IdP Sync
Identity Lifecycle and Offboarding
0/3
When an employee leaves, do you revoke access automatically?
Our HRIS or IdP is the source of truth for user records. Agent access inherits from it, not from a separate admin spreadsheet
When an employee is offboarded, their agent connections are revoked automatically on the next sync cycle. No manual cleanup required
When an employee changes role or department, their agent policies update to reflect the new attributes automatically
07 · Audit Log
Audit Trail and Compliance Readiness
0/4
Can you show an auditor exactly what every agent did, which policy fired and who approved?
Every agent action produces an immutable log entry recording what happened, who was behind it, which app was involved, which rule applied and what the outcome was
Audit events include request and response records so an auditor can reconstruct exactly what data the agent read or wrote
Our audit trail maps every agent action to the human who delegated it, creating an unbroken chain from employee to agent to app
We can filter audit events by user, agent, app, action type and outcome, then export for SIEM ingestion or compliance reporting
See your results
Enter your work email to see your full governance assessment, score and actionable recommendations.
By submitting this form, you agree to receive other communications from Cakewalk per our Privacy Policy. Learn how we store, process and use your data.
AI Agent Governance Gap Analysis
cakewalk.security
0/25
Ungoverned
Want an engineer to walk through your gaps?
One click. We already have your email. An engineer will reach out to scope what it takes to close your open dimensions.
Save your results
Download a PDF or share a link that preserves your exact assessment
