Who collects the keys when an AI agent retires?
collects disclosed that a credential was leaked, exposing a trust relationship that could have been abused.
I’ve started describing agents as vendor employees who never sleep and never get offboarded.
Usually it’s a checklist built up over years of people leaving: laptop, badge, email, Slack, the company card, that one design tool only three people use.
It may have scheduled a job for next week or passed half a task to another agent; say one keeps a customer record current, and a renewal system reads that record every Monday.
Two years later, a customer’s lawyer asks why the old agent denied her application, and so take the exit checklist you already run for people and go through it one step at a time, asking what the agent version looks like.
Cached results and vector database entries can hold customer data, and some of it has to be kept for audit while the rest can go the same day.
The system moves through a formal offboarding stage that can require the owner’s sign-off and a decision on its data, and every step is logged as evidence for the auditor.
Their manager reassigned the open tickets, and HR filed their records somewhere they’ll stay for years, in case anyone ever challenges a decision they made.
The reasons sit in what was true that day: which model and prompt version ran, the data it saw, the tools it called and the policy in force; it held credentials and made decisions someone may ask about years from now.
For agents, I’d make that record answer two questions, each with a timestamp: when was the agent’s identity disabled, and when did anything last act under it?
Give every agent a named owner and a review date when it goes live, the same way a new hire gets a manager and a probation review.
What happened
I’ve started describing agents as vendor employees who never sleep and never get offboarded.
Usually it’s a checklist built up over years of people leaving: laptop, badge, email, Slack, the company card, that one design tool only three people use.
What changed
It may have scheduled a job for next week or passed half a task to another agent; say one keeps a customer record current, and a renewal system reads that record every Monday.
Who is affected
Two years later, a customer’s lawyer asks why the old agent denied her application, and so take the exit checklist you already run for people and go through it one step at a time, asking what the agent version looks like.
Why it matters
Cached results and vector database entries can hold customer data, and some of it has to be kept for audit while the rest can go the same day.
The system moves through a formal offboarding stage that can require the owner’s sign-off and a decision on its data, and every step is logged as evidence for the auditor.
Technical details
Their manager reassigned the open tickets, and HR filed their records somewhere they’ll stay for years, in case anyone ever challenges a decision they made.
The reasons sit in what was true that day: which model and prompt version ran, the data it saw, the tools it called and the policy in force; it held credentials and made decisions someone may ask about years from now.
Response
For agents, I’d make that record answer two questions, each with a timestamp: when was the agent’s identity disabled, and when did anything last act under it?
Give every agent a named owner and a review date when it goes live, the same way a new hire gets a manager and a probation review.
What security teams should do
The record should also name who owned the agent and where the evidence for its decisions now lives.