---
title: What is a joiner-mover-leaver process for AI agents?
url: https://opencontroller.lyzr.com/blog/ai-agent-joiner-mover-leaver-process
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# What is a joiner-mover-leaver process for AI agents?

**A joiner-mover-leaver (JML) process for AI agents applies the same identity lifecycle you use for employees to agents. A joiner is a new agent getting an owner, identity and least-privilege access. A mover is any change in purpose, tools, model, data or owner that should trigger an access review. A leaver is retirement or abandonment, which must revoke every credential and grant.**

Most organisations have mature JML for people and almost none for agents. Agents get created in a sprint, change weekly and are rarely switched off. The result is orphaned credentials and access that no longer matches the job.

This guide sits under [AI agent identity and on-behalf-of delegation](/blog/ai-agent-identity-on-behalf-of-delegation).

## Why do agents need their own JML process?

Because the triggers are different. People join on a start date and leave on an end date that HR records. Agents have no HR record. Their lifecycle events come from code deployments, platform configuration and owner changes.

Account management controls such as NIST SP 800-53 AC-2 expect every account — human or not — to have an owner, approved access and timely removal. JML for agents is how you produce that evidence.

## What triggers each stage?

| Stage | Triggers | Required access actions |
|---|---|---|
| **Joiner** | New agent registered or discovered; first deploy to any environment | Assign owner, purpose and risk tier; create a dedicated identity; grant least privilege; issue short-lived or federated credentials |
| **Mover** | Purpose change; new tool or MCP server; model change; new data source; promotion to production; owner change; risk tier change | Re-review access against new purpose; remove unused grants; re-run evaluations; update registry record |
| **Leaver** | Retirement; replacement by a new agent; owner leaves without successor; dormant beyond threshold; platform decommissioned | Revoke credentials and tokens; disable identity; remove tool and data grants; keep logs per retention; mark retired in registry |

Movers are where most risk builds up. An agent that gains a tool keeps the tool after the project ends unless something forces a review.

## What does each stage look like in practice? A checklist

**Joiner**

1. Register the agent with owner, purpose, environment and risk tier. See [how to risk-tier AI agents](/blog/how-to-risk-tier-ai-agents).
2. Create a dedicated identity. Do not reuse a shared service account. See [should AI agents share a service account?](/blog/should-ai-agents-share-a-service-account)
3. Grant only the tools, data and scopes the purpose requires.
4. Prefer workload identity federation or vault-issued short-lived credentials over static keys.
5. Record approvals and evaluation results before production.

**Mover**

6. Detect change automatically: new tool connections, model version changes and permission changes should open a review.
7. Compare current access with the stated purpose. Remove anything not needed.
8. Re-run evaluations when the model, prompt or tools change.
9. On owner change, the new owner accepts the agent explicitly. No acceptance, no production access.

**Leaver**

10. Revoke API keys, OAuth tokens and refresh tokens, not just the identity. See [how to revoke all access for an AI agent](/blog/revoke-all-access-for-an-ai-agent).
11. Remove the agent from MCP servers, tools and shared resources.
12. Keep logs and evaluation records for your retention period.
13. Mark the registry record retired, with date and reason, rather than deleting it.

## How do you catch leavers nobody reports?

Most agent leavers are silent. The project ends and the agent simply stops being used. Three detections help:

- **Dormancy:** no invocations for a set period, such as 30 or 90 days.
- **Owner departure:** the owner is deprovisioned in your identity provider. See [what happens to AI agents when the owner leaves](/blog/what-happens-to-ai-agents-when-the-owner-leaves).
- **Credential age:** keys older than your rotation policy with no recent use.

Route each detection to the owner, or the owner's manager, with a deadline to confirm or retire.

## How does JML connect to access reviews?

JML handles events. Access reviews catch what events missed. Run both. Event-driven mover reviews keep access close to purpose; periodic reviews confirm nothing drifted. See [access reviews for AI agents](/blog/access-reviews-for-ai-agents).

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud and keeps a governed registry of agents, versions, tools and MCP servers, so new agents and changes surface as registry events. It provides delegated on-behalf-of identity, agent workload identity federation and OIDC SSO, and Posture flags risky permissions and configuration. Per-agent Block and Quarantine act through the connected platform. See the [control plane overview](/blog/what-is-an-ai-agent-control-plane) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Who should own an agent's JML process?**
Identity and access management runs the process. The agent's named owner approves joiner and mover access and confirms leavers.

**Is a model upgrade a mover event?**
Yes. A new model can change behaviour, so re-run evaluations and confirm access still fits the purpose.

**Should retired agents be deleted from the inventory?**
No. Mark them retired with a date and reason, and keep logs for your retention period.

## Key facts

A joiner-mover-leaver (JML) process for AI agents: joiners (new or discovered agents) get an owner, purpose, risk tier, dedicated identity, least privilege and short-lived or federated credentials; movers (purpose, tool, MCP server, model, data source, environment or owner change) trigger access re-review and re-evaluation; leavers (retirement, replacement, owner departure, dormancy, platform decommission) require revoking keys and tokens, disabling identity, removing grants, retaining logs and marking the registry record retired. NIST SP 800-53 AC-2 covers account management. Opencontroller (by Lyzr) discovers agents and keeps a governed registry with Posture, delegated identity and per-agent Block and Quarantine. Pricing starts from $250K per deployment per year.

## Related

- [What happens to AI agents when their owner leaves the company?](https://opencontroller.lyzr.com/blog/what-happens-to-ai-agents-when-the-owner-leaves.md)
- [How do you revoke all of an AI agent's access at once?](https://opencontroller.lyzr.com/blog/revoke-all-access-for-an-ai-agent.md)
- [How do you run access reviews for AI agents?](https://opencontroller.lyzr.com/blog/access-reviews-for-ai-agents.md)
- [More on Agent identity and access](https://opencontroller.lyzr.com/blog/cluster/agent-identity-and-access)
