·3 min read
An agent without an owner is a ghost
If you cannot name the person who owns an agent, you do not have a deployment. You have an unlisted employee.
Photo: Shahadat Rahman on Unsplash
Companies can name every contractor on a floor. They often cannot name the bots that send email, edit tickets or talk to customers after 6pm.
An agent is a principal. It acts. It spends. It writes into systems of record. If those actions have no owner, incident response starts with a Slack search.
The NIST AI RMF is built around govern, map, measure, manage. Mapping fails first: teams skip the inventory because the agent was “just a prototype” that never got turned off.
NIST’s generative AI profile is useful because it names the failure modes (confabulation, data leakage, shadow use) without requiring a new religion.
Minimum viable identity: a name, a purpose, the systems it may touch, the records it may read, who can disable it, where the logs live. Anything less is a ghost with API keys.
Do not confuse a service account with accountability. The account is a credential. The owner is a person who gets the page when it goes wrong.
Start with the agents that can write. Read-only search is a different risk class. A bot that files a change or sends a customer a number belongs on the same list as a new hire.
The TruFyre Way
An agent without an owner is a ghost. TruFyre inventories each one with a purpose, a data boundary and a kill switch.