The operational problem
When AI takes an action, the question that determines liability is not what did the AI do? but who authorised that action class, in advance, in writing, with named accountability? The answer determines whether the firm carries the action as institutional governance or as accountability dilution.
In most legal functions today the answer is implicit. The AI Use Policy (GOV-02) names the scope; the AI Task Force (STR-07) approved the procurement; the practice group operates the tool. Nowhere is there a written delegation of authority that specifies what the system may decide, within what scope, and which human is accountable when the decision is wrong.
That gap is Class 9 (Accountability dilution) operationalised. Without an explicit Delegation-Authority Register, every Tier 3 (Workflow operator) and Tier 4 (Autonomous agent) capability deployment carries the dilution risk by default. The function cannot answer the regulator’s first question about an autonomous action — who delegated this authority? — because no document holds the delegation record.
Legal AI OS Relevance
The Delegation-Authority Register Architecture sits at the intersection of Pillar P4 (Governance, Risk & Defensible AI) and the Governance Layer (Layer G). It produces evidence for two Defensibility Elements:
- DE-1 Decision traceability — the register specifies which decisions the system may make and therefore the trace of who authorised the decision class
- DE-4 Governance posture — the signed register is itself the governance-posture artefact at the level of the system, not the transaction
Anchoring to Bands 3, 4, and 5 (Integrated → Optimised → Defensible), the Module is required before any Tier 3+ capability is deployed at Band 3 or above. At Band 4 the register is operated at quarterly cadence with the Materiality Calibration (GOV-16) refresh. At Band 5 the register is integrated with the Evidence Register (GOV-13) such that every delegated decision class has a traceable evidence row. Methodology v2026.1.
Pillar Alignment
Pillar 4 (Governance, Risk & Defensible AI) is the institutional capability that holds AI accountable. The Delegation-Authority Register operationalises accountability at the system level — what the system may decide, under what scope, with which human held accountable when the decision is wrong.
The Pillar 4 posture this Module advances: from implicit authorisation through procurement approval to explicit, signed, named, scope-bounded delegation per Tier 3+ capability.
Operating Layer Impact
The Governance Layer (Layer G) is the institutional substrate that constrains how AI may be operated. The Delegation-Authority Register is the Layer-G artefact that names the accountability for every Tier 3+ capability. The Agentic Governance Charter (GOV-08) writes against the register. The Evidence Register (GOV-13) writes against it. The Materiality Calibration (GOV-16) tunes the thresholds the register specifies.
Without the Delegation-Authority Register, the Execution Layer operates AI under implicit authority and the function cannot defend the autonomy posture at any cadence.
Maturity Band Relevance
The Module strengthens the Defensibility and Autonomy lenses of the Maturity Stack.
From Band
To Band
Defensibility-lens shift
Autonomy-lens shift
3 — Integrated
4 — Optimised
Implicit authorisation → explicit signed register per Tier 3+ capability
Tier 3 deployed without delegation → Tier 3 deployed under documented delegation
4 — Optimised
5 — Defensible
Register operated at quarterly cadence with materiality calibration
Tier 4 (autonomous agent) deployment is gated by register population
5 — Defensible
(sustaining)
Register integrated with Evidence Register; per-decision audit trail
Autonomy posture defensible to regulator at the level of the system
Tier 3 (Workflow operator) capability requires the register to deploy. Tier 4 (Autonomous agent) capability requires the register and the Agentic Governance Charter (GOV-08) and the Evidence Register (GOV-13) and the Materiality Calibration (GOV-16) in operation.
Operational Outcomes
Operating the Architecture produces four institutional artefacts:
Artefact
Purpose
DPS evidence stream
Delegation-Authority Register — per Tier 3+ capability
The signed record of what the system may decide and which human is accountable
DE-1 + DE-4 (primary)
DPS Governance Posture section input
The register feeds Element 4 directly
DE-4
Escalation path documentation
Per delegation entry, the path when scope is exceeded or guardrail is hit
DE-1
Sunset archive record per decommissioned capability
At capability retirement, the register entry archives with the final post-mortem
DE-4
Records retain for the regulator’s limitation period for the longest applicable jurisdiction, plus the full lifecycle of the AI system.
Defensibility and Governance Considerations
The Module produces evidence for DE-1 (Decision traceability) and DE-4 (Governance posture). The Risk Class addressed is Class 9 (Accountability dilution); the register’s existence is the canonical mitigation for Class 9 at the system level.
For Tier 4 (Autonomous agent) capability, the register requires:
- A delegation entry per decision class the system may execute autonomously
- Linkage to the Materiality Calibration (GOV-16) row for each in-scope decision type
- An escalation path: what triggers human handoff, who handles, on what cadence
- A revocation trigger: which performance or risk condition revokes the delegation
- A signature by the GC or delegated AI Task Force Chair, dated, with effective date
For Tier 3 (Workflow operator) capability, the register requires:
- A delegation entry per workflow the system operates
- The boundary: what the system may not do without exception-triggered HITL per GOV-16
- The accountable individual: by named role and named person
- The audit cadence: when the delegation is reviewed against actual operation
The editorial-independence attestation applies — the Module makes no vendor-specific recommendations.
Institutional Use Cases
Use case 1 — A 14-partner firm deploying first Tier 3 capability. The firm is at Band 3. The contract-extraction capability operates against a workflow that previously required associate review. The Delegation-Authority Register entry specifies: the delegated decision class is extract canonical clauses from a contract for partner review; the system may not decide legal sufficiency; the accountable human is the supervising partner per the engagement; the escalation path is the senior partner for any extracted clause flagged as ambiguous; the revocation trigger is more than two errors per ten extractions in a quarter. Signed by GC; effective dated 2026-Q3.
Use case 2 — A 200-lawyer in-house function deploying Tier 4 autonomous agent for vendor-renewal triage. The function is at Band 4. The capability triages incoming renewal requests against a pre-approved playbook, auto-approving routine renewals and routing non-routine to a named contracts lead. The register entry specifies: 18 decision classes (renewal-as-is up to threshold, renewal-with-price-change within band, etc.); 12 are in autonomous scope; 6 require exception-triggered HITL per GOV-16. The Materiality Calibration row specifies the financial threshold at which Full HITL is mandatory regardless of class. The Agentic Governance Charter (GOV-08) chairs the standing review. The Evidence Register (GOV-13) captures every autonomous decision for sample audit. The capability deploys only when all four artefacts are operational and signed.
Use case 3 — A global energy GC office at Band 5 with 11 Tier 3+ capabilities. The register holds 11 entries with collectively 137 in-scope decision classes. The annual external attestation samples 22 decisions across the year and verifies each against the register entry. The attestation finding informs the register’s annual refresh; one delegation revoked in the year due to performance degradation; two new delegations added; one capability migrated from Tier 3 to Tier 4 with the corresponding Agentic Governance Charter activation.
Recommended Stakeholders
RACI role
Stakeholders
Owner
General Counsel
Approvers
General Counsel · AI Task Force Chair
Contributors
Legal Operations · Risk and Compliance
Informed
Board · Audit Committee · CIO / CISO
The GC holds the signing authority per delegation entry. The AI Task Force Chair (per STR-07) is the standing forum where new delegations are deliberated; the GC signs after Task Force recommendation.
Implementation Complexity
Dimension
Specification
First run per Tier 3+ capability
2 hours
Quarterly refresh per capability
30 minutes
Cross-team dependencies
Risk (escalation pathway design) · IT (system-actual-behaviour validation) · the Practice Group (decision-class taxonomy)
Self-serve viability
Partial — first run benefits from advisory at Tier 4; Tier 3 is self-serve
Advisory recommendation
Programme Design for first Tier 4 capability; Strategic Retainer for functions operating ≥3 Tier 3+ capabilities
The Module is methodology-versioned (v2026.1); each entry’s effective date and next review are the operating telemetry.
Inputs
Input
Source
Capability Portfolio Tier 3+ capability list
SUS-10
Materiality Calibration HITL tier per decision type
GOV-16
AI Use Policy delegated authority scope
GOV-02
Agentic Tier classification per capability
STR-07 + Agentic Governance Charter where applicable
Framework — the Delegation Entry
Entry structure
The canonical structure is one entry per Tier 3+ capability. Each entry holds:
Section
Content
Capability identifier
Canonical name from Capability Portfolio (SUS-10)
Tier classification
Tier 3 (Workflow operator) or Tier 4 (Autonomous agent)
Effective date
Date the delegation is in force
Next review
Quarterly cadence date
Accountable human
Named individual + role
Delegated decision classes
One row per class; what the system may decide
Guardrails
What the system may not decide without exception-triggered HITL
Materiality threshold
Linked to GOV-16 row
Escalation path
Trigger condition + recipient + handover protocol
Revocation trigger
Performance or risk condition that revokes the delegation
Signatory
GC or delegated AI Task Force Chair
Evidence Register cross-reference
GOV-13 row IDs
Decision class taxonomy
The canonical pattern: each decision class is specified at the verb level — recommend, triage, flag, route, summarise, extract, generate-draft, approve-routine, execute. Verb specificity prevents scope creep.
Verb
Tier 3 typical
Tier 4 typical
Recommend
Yes — system recommends; human decides
Yes — same
Triage
Yes — system routes per playbook; human handles
Yes — system handles routine per playbook; human handles flagged
Flag
Yes
Yes
Route
Yes
Yes
Summarise
Yes
Yes
Extract
Yes
Yes
Generate-draft
Yes — for human review
Yes
Approve-routine
No — Tier 4 only
Yes — within financial/scope threshold per GOV-16
Execute
No — Tier 4 only
Yes — within delegated authority
Guardrails
For each delegated decision class, guardrails specify what triggers exception-triggered HITL or full handover. Common guardrails:
Guardrail
Trigger
Materiality threshold
GOV-16 Full HITL boundary crossed (financial, regulatory, client-impact)
Confidence threshold
System’s own confidence score below the calibrated floor
Out-of-scope detection
Decision class not in the delegated taxonomy
Anomaly detection
Decision pattern deviates from the population in the audit sample
Regulatory change
New rule activates; pending delegation review
Escalation path
Trigger
Recipient
Handover protocol
Guardrail hit
Named accountable human
System pauses; package context to recipient; recipient decides
Revocation trigger met
AI Task Force Chair
Delegation revoked; capability returns to Tier 2 review or sunset
Regulatory inquiry on a delegated decision
GC
Evidence Register (GOV-13) row produced; delegation documented
Revocation triggers
Trigger
Action
Performance below the calibrated floor (e.g., > 2% error in audit sample)
Delegation revoked pending review
Audit finding requiring scope reduction
Delegation amended at next quarterly cadence
Regulatory change altering the scope
Delegation suspended pending GOV-16 recalibration
Capability sunset
Delegation archived with the Capability Portfolio sunset entry
Worked Example
A 200-lawyer in-house function at Band 3 deploys its first Tier 3 capability: a contract-extraction system that extracts canonical clauses for partner review.
Step
Activity
Output
1
Capability Portfolio entry confirms in-scope; Tier 3 classification
SUS-10 entry
2
Materiality Calibration row populated: extraction is Audit-only; flagging is Exception-triggered HITL
GOV-16 row
3
Delegation-Authority Register entry drafted: decision class list, guardrails, escalation path, revocation trigger
Entry v0.1
4
AI Task Force review at next monthly session
Recommended to GC
5
GC signs the entry
Entry v1.0, effective 2026-09-01
6
Evidence Register rows updated to cross-reference the delegation
GOV-13 updated
7
Capability deployed under delegated authority
Operating
Quarterly cadence at quarter +1: audit sample of 30 extractions; 1 error; well within revocation trigger threshold. Delegation continues. At quarter +4: annual review; performance trend reviewed; capability promoted to Tier 4 candidacy under GOV-08 evaluation.
Common Failure Modes
Failure mode
Detection signal
Recovery
Tier 3+ capability deployed without register entry
Capability live; register row absent
Pause capability; populate entry; re-deploy under delegation
Delegation entry signed but never refreshed
Last-review date older than two quarters
Bind refresh to Governance Cadence (GOV-15)
Guardrail trigger fires but system continues
Out-of-scope decision class executed
Capability paused; AI Task Force investigation; revocation considered
Accountable human named but not in role
Role turnover loses the accountability trace
Bind register updates to HR change-notice cadence
Materiality threshold drifts without recalibration
GOV-16 row not refreshed; threshold stale
Standing quarterly recalibration per GOV-16
Tier 3 → Tier 4 promotion without Agentic Charter
Capability operating at Tier 4 scope but Tier 3 register entry
Pause; activate GOV-08; rewrite delegation entry
Revocation trigger met but not actioned
Performance breach in audit; no register update
Audit Committee escalation; documented remediation plan
Edge Cases
The Module does not apply when:
- The function operates only Tier 1 (Standalone assistant) and Tier 2 (Workflow assistant) capabilities — no delegation entries required; the AI Use Policy (GOV-02) is sufficient
- The function operates under a parent organisation’s enterprise-wide delegation architecture — the Module produces sub-entries that feed the parent’s enterprise register
- The function is below Band 3 — Tier 3+ deployment is premature; build Bands 1-2 through CHG-01 + GOV-01 first
- The capability is third-party-hosted and the third party holds the delegation under a separate framework — the Module produces a counterparty-attestation entry rather than a delegation entry