What is an AI system owner?
An AI system owner is one named person accountable for what an AI system is for, which providers it uses, and whether its controls still fit. NIST AI RMF GOVERN 2.1 asks for documented roles and responsibilities, and EU AI Act Article 26(2) asks deployers of high-risk systems to assign human oversight to named people. An owner who has left the company is no owner, so ownership has to follow team membership automatically.
An AI system is an application or agent together with the model providers and tools it uses. The owner is a person, not a team. A team can share the work, but when a question about the system needs an answer, such as whether it may start reading customer records, one person gives it.
Ownership is the second of the four records that AI governance frameworks ask for. It comes after the AI inventory, because you can't own what isn't listed, and before provider decisions and reviews, which need someone to make them. For how the four fit together, see what is AI governance.
What is an AI system owner accountable for?
An AI system owner is accountable for keeping the system's record true and for acting when it stops being true. In practice, that's the following five responsibilities:
- Purpose. The owner writes what the system is for, and which data classes and tools it may use, and changes it when the use changes.
- Providers. The owner knows which model providers and MCP (Model Context Protocol) servers the system uses, and asks for a decision on any that aren't approved.
- Reviews. The owner checks the record at least every 180 days, and sooner when something changes. For how, see how to run periodic AI system reviews.
- Held actions. When a Guard policy holds an action for a person to decide, Arcjet routes it to the owner.
- Notices. The owner reads the weekly owner digest of overdue reviews and violations, and the drift notices for their systems, and acts on them.
The owner isn't necessarily the person who builds the system or the person who writes its runtime policies. The owner decides what should be true, engineers make it true, and the security team checks that it is.
How does an owner differ from an approver and an administrator?
An owner answers for one AI system, an approver signs off on a specific change, and an administrator maintains the records. The three are often confused because one person can hold more than one role. The following table separates them.
| Role | Scope | What they do | In Arcjet |
|---|---|---|---|
| Owner | One AI system, actor, or provider | Accountable for purpose, providers, and reviews; decides held actions | The named owner on the record. Receives the owner digest and Hold requests |
| Approver | One change | Confirms a change before it takes effect | A second approver for publishing a Guard policy |
| Administrator | The whole team | Assigns owners, edits purpose, and records provider decisions | A team administrator. Only administrators edit owner, purpose, and decisions; other members read them |
Keep the approver separate from the author of a change. When a second approver signs off on a Guard policy before it publishes, one person can't change what an agent is allowed to do on their own.
Which frameworks require a named AI owner?
The NIST AI RMF and the EU AI Act each have a requirement that a named owner supplies evidence for. The following table lists the references, with short titles in Arcjet's wording rather than the frameworks' text. The evidence page also maps owners to ISO/IEC 42001, which the table leaves out because the standard's text is sold rather than published openly.
| Framework | Reference | What it asks for |
|---|---|---|
| NIST AI Risk Management Framework (AI RMF) 1.0 | GOVERN 2.1 | Roles and responsibilities for AI risk are documented |
| EU AI Act | Article 26(2) | Deployers of high-risk systems assign human oversight to people with the competence, training, and authority to do it |
A named owner is evidence for each of these, but not all of it. GOVERN 2.1 also covers lines of communication, and Article 26(2) asks that the person has the competence, training, authority, and support, which is a judgment that a record can't make. For the primary texts, see the NIST AI RMF, Regulation (EU) 2024/1689 on EUR-Lex, and ISO/IEC 42001 on iso.org.
What happens when an AI system owner leaves?
When an owner leaves your Arcjet team, the system reads as having no owner, automatically. Nobody has to find and clear the old name. In Arcjet, an owner must be a member of the team, so removing the member removes the ownership, and the system appears under the No owner filter.
This matters more than it seems. A spreadsheet keeps a leaver's name in the owner column for months, and every report built from it shows full ownership. An auditor who checks one name against the staff list finds the gap, and then doubts every other row. A record that lapses on its own shows the true state: ownership is missing, and someone needs to pick it up.
Reassign ownership as part of offboarding. When you know someone is leaving, ask them to list their systems with the Owned by me filter, and have an administrator move each one to a new owner before their last day. The weekly owner digest goes only to current owners, so an unowned system gets no reminders until someone takes it.
What rights does an AI system owner need?
An AI system owner needs held actions routed to them, read access to everything about the system, and the notices that say something changed. Arcjet gives owners the following:
- Hold decisions. When a Guard policy holds an action, Arcjet routes it to the system's owner.
- The owner digest. A weekly email lists violations on the owner's providers and the owner's records that are overdue for review. It sends nothing when nothing needs attention.
- Drift notices. When Arcjet observes a data class or tool that the purpose doesn't declare, the owner gets a notice. For how drift works, see AI purpose drift.
- Read access. Every team member reads the owner, purpose, decisions, and review dates.
Owners don't need administrator rights to do their job. Team administrators edit the owner, purpose, and provider decisions. Keeping those edits with administrators means each change has a small set of authors, and every change is written to the WorkOS audit log with the fields that changed and the new owner. The purpose text itself isn't written to the log.
What does an AI governance RACI look like?
A RACI (responsible, accountable, consulted, informed) table assigns each governance task to roles. The following template fits a team with a security function and engineering teams that build or run AI systems. Adapt the roles to your organization.
| Task | Owner | Engineering | Security | Administrator |
|---|---|---|---|---|
| Confirm the AI system in the inventory | A | R | C | I |
| Write the purpose and declared data classes | A | C | C | R |
| Decide on a provider or client | A | C | C | R |
| Write and test runtime policies | C | R | A | I |
| Approve a policy for publication | I | C | R, A | I |
| Decide a held action | R, A | I | I | I |
| Review the system every 180 days | R, A | C | C | I |
| Export evidence for audit | I | I | R, A | C |
In the table, R is the role that does the work, A is the one person who answers for it, C is consulted before, and I is informed after. Where the administrator is R for a record, the owner decides and the administrator enters it.
How do you assign owners to AI systems, step by step?
The following procedure assigns an owner to every AI system and keeps ownership current.
- Find systems with no owner. In the inventory, filter by No owner. Include actors and providers as well as confirmed AI systems.
- Choose one person per system. Pick the person who can answer for the system's purpose, usually the product or engineering lead for a system you build, or the head of the team that uses a tool. Choose a member of your Arcjet team.
- Record the owner and purpose. A team administrator sets the owner and writes the purpose, with the data classes and tools the system may use.
- Tell the owner what they own. Share this page and the RACI table, and confirm the owner receives the weekly digest.
- Mark the system reviewed. The first review confirms the record is complete and starts the 180-day cycle.
- Reassign during offboarding. When an owner leaves, systems they owned show No owner. Assign each one a new owner in the same week.
Where Arcjet fits
Arcjet puts the owner on the observed record: the AI systems, actors, and providers in your AI inventory. An owner must be a team member and lapses when they leave, so the record can't show a stale name. Owners receive held actions, drift notices, and a weekly digest, which gives ownership a job to do.
On the evidence page, the share of systems with an owner supplies evidence for NIST AI RMF GOVERN 2.1 and EU AI Act Article 26(2), and the evidence page also maps it to ISO/IEC 42001. It doesn't show that an owner is competent, that they have the authority your organization needs, or how concerns are reported to them. Those stay with your governance process.
With owners in place, record provider decisions next. See how to approve or prohibit AI providers, and for the full governance picture, see what is AI governance.
Frequently asked questions
Can a team own an AI system instead of a person?
A team can share the work, but name one person as owner. When a question needs an answer, such as whether the system may start handling a new kind of data, one person gives it. In Arcjet, the owner is a member of your team, not a group.
Does an AI system owner need administrator rights?
No. Owners receive held actions, drift notices, and the weekly owner digest, and every team member reads the record. Team administrators edit the owner, purpose, and provider decisions, so changes have a small set of authors and each one is written to the WorkOS audit log.
Who should own an AI system you buy rather than build?
The head of the team that uses it, or the person who asked for it, is usually the right owner. They know what it's for and who uses it. Security and procurement are consulted on the provider decision, but the owner answers for the purpose.
Is a named owner enough for EU AI Act human oversight?
No. For high-risk systems, Article 26(2) asks deployers to assign oversight to people with the competence, training, and authority to carry it out. A named owner is evidence that someone is assigned. Competence and authority are judgments your organization makes and documents.
What does Arcjet log when an owner changes?
The WorkOS audit log records which governance fields changed and the new owner. It never records the purpose text. When an owner leaves the team, the system reads as having no owner without anyone editing it.
AI runtime security in your code
Protect your AI agent workflows with Arcjet
Get allow, deny, and redact on agent actions before the side effect.