1. Objective
To ensure that for every AI system an organization operates, a specific person is accountable for what it is for, what it can reach, what it costs and whether it should keep running.
2. Scope
Applies to AI systems used by or on behalf of the organization, including: AI agents and copilots built on low-code or pro-code platforms; automated flows that invoke AI models; direct model and API connections; AI tool and integration servers (for example Model Context Protocol servers); and non-human identities (service principals, API keys, tokens) used by AI systems.
3. Key terms
Owner of record. The single human, resolvable to an active identity in the organization's directory, accountable for an AI system.
Creator. The identity that built or published the system. A creator is not presumed to be the owner of record.
Ownership roles. Business, technical, security, data and executive owners recorded for higher-risk systems.
Owner unknown. The state of a system for which neither the platform nor the organization can identify an accountable person.
4. Requirements
5. Evidence
- Inventory export showing owner of record, ownership roles and last attestation per system.
- Leaver and mover reassignment log with dates.
- List of "owner unknown" and ownerless systems with escalation records.
- Audit log of manual ownership changes.
6. Metrics
| Metric | Definition |
|---|---|
| Ownership coverage | Systems with a current owner of record, divided by all in-scope systems. |
| Owner-unknown share | Systems for which no owner can be identified, divided by all in-scope systems. |
| Reassignment time | Median days from leaver event to reassignment. |
| Attestation currency | Systems with an attestation less than 12 months old, divided by owned systems. |
7. Basis for conclusions
Field experience shows that a large share of low-code agents in some estates have no owner reported by the platform, and that the people who created agents frequently change roles. Defaulting such agents to an assumed owner creates false accountability. The draft therefore requires an explicit "owner unknown" state and a reassignment clock rather than inference.
This founding draft was prepared by the Secretariat from practical control experience, including the founding sponsor's, and is published for open comment. It names no product and requires none.
8. Questions for respondents
- Is 30 days an appropriate reassignment period? Should it vary by risk?
- Should every system require the five ownership roles, or only higher-risk systems as proposed?
- Is annual re-attestation sufficient for autonomous agents?