Joiner, mover, leaver (JML): the identity lifecycle for people and AI agents

Updated

Joiner, mover, leaver (JML) is the identity lifecycle process that grants access when someone starts, changes it when they change role, and removes it when they leave. A joiner gets the access their role needs on day one. A mover gets the access of the new role and loses what only the old role needed, and a leaver loses everything, on time, in every system. Most access risk in an organisation traces back to one of those three going wrong.

Why JML matters

Access that is never removed accumulates. Someone who has changed team three times can end up holding the access of all four roles, and an account left active after its owner has gone is one nobody is watching. Auditors know this, which is why they test JML directly: they pick a sample of leavers and ask for evidence of when access was removed.

The failures tend to look the same everywhere. New starters wait days for access, so their managers ask for broad access just to get it done. Movers pick up access for the new role while nothing from the old one is taken away. Leavers have their directory account disabled, but the SaaS accounts and API keys they used stay live.

The joiner process

A joiner event usually starts in the HR system when a hire is recorded with a start date. Ideally the identity is created from that record before the start date, so that access is ready on day one.

The starting access should come from birthright rules, meaning the baseline every person in a given department, location or job gets. Accounts are then provisioned in each application, through a connector where one exists and through a ticket where it does not. Where the baseline includes something sensitive, the manager approves it first. Finally, record what was granted and why, so the first access review has something to compare against.

The anti-pattern is “copy the access of someone in the same team”, because you copy their accumulated access along with what the role actually needs.

The mover process

Movers are the hardest of the three because the right answer is to add and remove at the same time, selectively.

  1. Detect the move from a department, title, manager or location change in HR.
  2. Grant the new role’s access.
  3. Find the stranded access: what the person holds only because of the old role, and that the new role does not give them.
  4. Remove or review it. Low-risk stranded access can often go automatically, while anything sensitive should go to a person.
  5. Leave requested access alone, since access someone asked for and had approved is not stranded just because their department changed.
  6. Trigger a review of the person’s access if the move is a significant one.

The leaver process

The leaver is the event that has to be complete. It should run on the termination date, or immediately for an involuntary exit, and disable the person’s account in every connected application, with a ticket for each application that has no connector. Live sessions need ending too, otherwise an existing sign-in can outlast the account.

Then there is everything the person owns, such as service accounts, API keys and, increasingly, AI agents. Mail needs attention as well, because a forwarding rule can keep sending data out after someone has left. Throughout, the process should leave a record of each step, including what failed and what was handed to a person to finish.

Automating JML

Most JML automation has the same shape. HR is the source of truth for who works here, in which role and for what dates, and the identity system listens for HR events through a scheduled sync or through an API call from the HR system or its middleware.

Rules rather than tickets decide birthright access and what a mover loses. Connectors do the provisioning where they exist, and where they do not, the step becomes a tracked ticket so nothing is silently skipped. Because HR systems resend events, each event also needs an idempotency key, so that a resend does not create a duplicate.

People stay involved for the risky decisions. In most teams that means approving sensitive joiner access, deciding high-risk mover revocations and reassigning what a leaver owned.

What changes when people own AI agents

Your colleagues’ AI agents act with real permissions, reading a CRM or opening pull requests. Each agent has an owner, and the owner’s lifecycle now has to reach the agent.

When the owner leaves, their agents should not keep running with nobody accountable. The safe options are to deactivate the agent or, if a team still relies on it, keep it running and assign a new owner. Whichever you choose, it should be decided during the leaver rather than discovered months later. It also helps if whoever runs the leaver can see how many agents the person owns before they start.

Moves are less clear-cut. When the owner changes team, the agent may no longer belong with them, so review its access alongside theirs and move its ownership if the work stays behind. A leave of absence raises a similar question: if the person is away, an agent acting for them probably should not keep acting.

The same thinking applies to service accounts and API keys a person created. These are the identities most often left behind, because no HR event mentions them.

How AtmanID approaches it

In AtmanID, joiner, mover and leaver events arrive from HR feeds through Identity Sources, from the HRIS API (/api/v1/jml/joiner, /mover, /leaver), or are started from the JML page (Handle joiner, mover and leaver events). The API takes an idempotency key, and the same key within 24 hours returns the original result instead of creating anything (JML lifecycle tutorial).

The joiner workflow works out starting access from Birthright Access Profiles. It can wait for manager approval before provisioning applications, and it creates accounts through connectors or opens a ticket where it cannot. A hire sent through the API with a future start date is created as a pre-hire and cannot sign in until that date.

The mover compares old and new birthright access. Leftover low-risk access is revoked automatically and anything else becomes a revocation ticket, and access the person requested and had approved is never in that set.

The leaver disables each app account through its connector or opens an urgent revocation ticket, sets the person to deprovisioned, and suspends live sessions when session enforcement is on. For Exchange Online it also removes inbox forwarding rules and group memberships. A scheduled leaver runs on the termination date.

For agents, the Leaver form shows how many AI agents the person owns before you trigger it. The workflow flags the machine identities they own as orphaned and deactivates the AI agents they own, while agents with a managing team keep running and are flagged for a new owner. Pause puts a person on leave and suspends the AI agents they own, except during a secondment, and Return restores only the agents the pause suspended.

Every step writes to the audit trail, for example JML_LEAVER_TRIGGERED and AGENT_DEACTIVATED_ON_LEAVER, so you can answer an auditor’s leaver sample with a query.

FAQ

What is the JML process in IAM?

It is the process that keeps access in line with employment: provisioning access for joiners, adjusting it for movers and removing it for leavers. It is usually driven by HR events and carried out by an identity governance system through connectors and tickets.

What is the difference between a mover and a leaver?

A mover stays with the organisation but changes role, department or manager, so some access is added and some removed. A leaver departs, so all their access is removed and anything they own is reassigned or shut down.

What is birthright access?

It is the baseline access a person gets automatically because of attributes such as department or job title. Defining it as rules is what makes joiners fast, and it is also what makes mover clean-up possible.

How quickly should a leaver’s access be removed?

On the termination date, or immediately for an involuntary exit. The time between HR recording the departure and the last credential being revoked is the risky window, so measure it and work to keep it short.

What happens to a leaver’s AI agents?

That should be decided as part of the leaver process. You can deactivate them, or reassign them to a new accountable owner when a team still needs them; leaving them running with no owner should not be an option.

Related: What is a non-human identity? · How to govern AI agents · AI agent access control