The Delegation-Authority Register is a governance mechanism that maps every Tier 3 (Workflow operator) and Tier 4 (Autonomous agent) AI capability to a clearly named human accountable owner. It is the primary control used to mitigate Class 9 (Accountability dilution) at the per-capability level.
For each Tier 3+ capability, the Register records:
- Delegated decision classes: the specific types of decisions the AI system is allowed to make.
- Guardrails: decisions or actions the system is not permitted to take without triggering an exception and human-in-the-loop (HITL) review.
- Materiality threshold (linked to GOV-16): the impact level at which decisions become materially significant and may require additional controls or oversight.
- Escalation path: who and how to escalate when decisions exceed scope, risk, or materiality thresholds.
- Revocation trigger: conditions under which the delegation is suspended or revoked.
- Signatory: the accountable approver (General Counsel or a delegated AI Task Force Chair) who formally authorises the delegation.
The Delegation-Authority Register is operated at corpus scale by the Delegation-Authority Register Architecture (GOV-14) module. This module ensures that every Tier 3 and Tier 4 capability has:
- Explicitly defined decision scope and constraints.
- A named human accountable for errors or harms arising from delegated decisions.
- A structured escalation and revocation mechanism.
Operational cadence and dependencies
- The Register is refreshed quarterly in line with the Governance Operating Cadence (GOV-15), ensuring that changes in systems, risk posture, or organisational structure are reflected in the delegations.
- Tier 4 (Autonomous agent) capabilities require additional coverage under the Agentic Governance Charter (GOV-08), adding stricter conditions and oversight for higher-autonomy systems.
Evidence and assurance outputs
The Register feeds into the Decision & Posture Store (DPS) by producing:
- DE-1 (Decision traceability) evidence: showing which system made which decision, under which delegation, and who the accountable human is.
- DE-4 (Governance posture) evidence: demonstrating the organisation’s current delegation landscape, coverage of Tier 3/4 systems, and alignment with governance policies.
In practice, the Delegation-Authority Register ensures that no Tier 3+ AI capability operates without:
- A clearly bounded decision mandate.
- Non-negotiable guardrails and HITL triggers.
- A documented, named human accountable for outcomes.
- A defined path to escalate or revoke authority when risk or performance changes.
Use the Delegation-Authority Register as the single source of truth for: (1) which AI systems are allowed to decide what, (2) who is accountable when those decisions go wrong, and (3) how to escalate or revoke that authority when risk thresholds are crossed.