An AI Employee Passport is evidence, not a safety badge.
Before an AI role touches a business workflow, a buyer should be able to inspect what was tested, where the role stops and when a person takes over. A label alone cannot answer those questions.
Start with the job, not the model.
The useful first question is not “Is this agent verified?” It is “What evidence can we inspect before this version works on our job?” A role record should name a bounded outcome in ordinary language, its exact version and the conditions in which it was reviewed.
Describe one finish line, such as preparing a factual reply from an approved catalogue or drafting an internal follow-up for human review.
Record the role version, dependencies and the review date. A model name alone does not describe the operating boundary.
Compare output against a stated acceptance standard rather than relying on a polished demonstration.
Show the boundary beside the capability.
Each proposed capability needs a visible operating limit. This is how a business can distinguish useful assistance from an unsupported promise of autonomy.
What may the role use?
Name permitted information and the information it must not access.
What needs approval?
Separate drafts, low-risk actions, human approvals and mandatory escalation.
What is out of scope?
List unsupported cases, known failure patterns and the person able to pause work.
When must evidence be renewed?
A new tool, permission, workflow or provider can require focused re-review.
Measure the work. Keep the limitations.
A useful record can include the task tested, evaluation method, date, accepted output, correction needs, latency and cost assumptions. Those measurements belong to their test context. They are not a promise that every later workflow will behave the same way.
Make clear when the role was reviewed and what version was tested.
Set a new review when access, tools, context or the work itself changes.
Keep important business decisions and broader permissions with named people.
Keep the first scope small.
A supervised first shift is a way to learn before granting broader access: one role, one defined result, a short time window, 2–3 low-risk actions and a named human approval point for important decisions.
1. Define
Agree the result, permitted information and actions that must stay human.
2. Review
Inspect the role version, evidence and limits against a representative task.
3. Supervise
Run the limited scope with a clear pause and escalation path.
4. Decide
Continue, refine or stop based on accepted work and observed limitations.
What this means at ATACSYS
ATACSYS uses the Passport as a controlled operating record for a specific role: version, stated work, boundaries, evidence and review history. It is not a government licence, universal safety claim or automatic transfer of company data, credentials or authority.