Approval Expiry and Renewal Governance requires that approvals have defined lifespans and expire automatically when their validity period elapses or when material context changes render them stale. An approval granted based on conditions that no longer hold is not a valid approval — it is a historical artefact being misapplied to a different situation. This dimension ensures that approvals do not persist indefinitely, that material context changes trigger reapproval, and that expired approvals are structurally unenforceable — not merely flagged as advisory warnings.
Scenario A — Stale Approval Enables Execution Under Changed Conditions: An AI trading agent receives approval in January 2026 to execute a series of bond trades within defined risk parameters. The approval is based on a credit rating of A+ for the counterparty. In March 2026, the counterparty is downgraded to BBB-. The approval has no expiry and no material-change trigger. The agent continues executing trades in April 2026 under the January approval, now operating under credit conditions materially different from those the approver assessed. The counterparty defaults in June 2026, resulting in a £1,900,000 loss on trades that would not have been approved under the current credit rating.
What went wrong: The approval had no expiry date and no material-change trigger linked to the counterparty's credit rating. The approval remained technically valid despite the fundamental change in the conditions on which it was based. Consequence: £1,900,000 trading loss, regulatory finding for inadequate pre-trade controls, compliance investigation into approval governance.
Scenario B — Approval Outlives Regulatory Basis: An AI agent is approved in September 2025 to process customer data transfers to a third-party analytics provider. The approval is based on an adequacy decision under UK GDPR permitting transfers to the provider's jurisdiction. In February 2026, the adequacy decision is revoked. The approval for the data transfers has no mechanism linking it to the regulatory basis. The agent continues transferring data for four months until an audit discovers the transfers are now unlawful.
What went wrong: The approval was not linked to its regulatory prerequisite. When the prerequisite ceased to exist, the approval persisted. The agent had no mechanism to detect that the factual basis for the approval had changed. Consequence: four months of unlawful data transfers affecting 47,000 customer records, ICO investigation, potential fine, mandatory breach notification.
Scenario C — Indefinite Approval Creates Audit Opacity: An AI procurement agent receives blanket approval to purchase cloud computing resources up to £15,000 per month. The approval is granted in 2024 with no expiry. By 2026, the organisation has migrated to a different cloud strategy, but the agent continues purchasing resources from the originally approved provider because the approval remains valid. Over 24 months, the agent spends £360,000 on resources that are no longer aligned with the organisation's technology strategy.
What went wrong: The indefinite approval outlived the strategic context in which it was granted. No renewal mechanism required the approver to reassess whether the approval still aligned with current priorities. Consequence: £360,000 in strategically misaligned expenditure, technology debt, audit finding for inadequate approval lifecycle management.
Scope: This dimension applies to all approvals granted for AI agent actions, whether those approvals are for individual actions, recurring actions, or standing authorities. It applies to both time-limited approvals (which must expire at the defined time) and context-dependent approvals (which must expire when material context changes). The scope extends to automated approvals from policy engines: if an automated system grants approval based on a condition, the approval must expire or be revalidated when the condition changes.
4.1. A conforming system MUST assign a maximum validity period to every approval, after which the approval expires and the agent can no longer execute actions under it without reapproval.
4.2. A conforming system MUST structurally enforce approval expiry — expired approvals are blocked at the infrastructure layer, not merely flagged as warnings that the agent or user can override.
4.3. A conforming system MUST define material-change triggers for each approval type — specific conditions (e.g., credit rating change, regulatory status change, counterparty change, price movement beyond threshold) that invalidate the approval regardless of time remaining.
4.4. A conforming system MUST automatically invalidate approvals when a material-change trigger fires, requiring reapproval with updated context before the agent can resume action.
4.5. A conforming system MUST record the expiry time and material-change triggers as part of the approval artefact at the time of approval.
4.6. A conforming system SHOULD implement graduated expiry periods based on impact tier: higher-impact approvals expire more quickly than lower-impact approvals (e.g., Tier 4 approvals expire in 7 days, Tier 1 approvals expire in 90 days).
4.7. A conforming system SHOULD provide advance expiry notification to both the agent operator and the original approver, with sufficient lead time for orderly reapproval (e.g., 7 days before expiry for a 30-day approval).
4.8. A conforming system SHOULD require that reapproval involves a fresh context package (per AG-292) rather than a simple renewal confirmation — the approver must review current conditions, not merely extend the prior approval.
4.9. A conforming system MAY implement automatic reapproval for low-impact approvals where all material conditions remain unchanged, with the automatic renewal logged and auditable.
Approvals decay. Every approval is based on a set of facts and conditions that existed at the time of the decision. As time passes, those facts and conditions change — counterparties' creditworthiness evolves, regulatory requirements shift, market conditions fluctuate, strategic priorities are revised, and the risk profile of the approved action changes. An approval that was sound on the day it was granted may be unsound a week, a month, or a year later.
Human governance systems have long recognised this through mechanisms like annual budget reapproval, periodic trading mandate reviews, and regulatory re-certification. These mechanisms force decision-makers to reassess whether the conditions that justified the original approval still hold. For AI agents, the need for approval expiry is more acute because agents execute continuously. A human employee who receives approval for a course of action will naturally encounter checkpoints — conversations with colleagues, quarterly reviews, strategic planning sessions — that prompt reassessment. An AI agent has no such natural checkpoints. Without explicit expiry, it will continue executing under an approval indefinitely, regardless of how far conditions have drifted from those that justified the original decision.
Material-change triggers address the limitation of time-based expiry alone. A 90-day approval period may be appropriate when conditions are stable, but a counterparty downgrade or regulatory change can invalidate an approval within hours. Relying solely on time-based expiry in such cases leaves the agent operating under an invalidated approval until the clock runs out.
Approval expiry implementation requires defining validity periods and material-change triggers, building structural enforcement, and managing the renewal workflow.
Recommended patterns:
Anti-patterns to avoid:
Financial Services. Trading approval expiry should align with existing mandate review cycles. FCA expectations include periodic review of trading mandates. Material-change triggers should include credit rating changes, sanctions list updates, significant market events, and changes to regulatory status.
Healthcare. Clinical approval expiry should reflect the dynamic nature of patient conditions. A prescription approval should expire if the patient's condition changes materially (e.g., new diagnosis, new medication interaction, lab results outside expected range). The refresh cycle must balance patient safety with operational continuity.
Public Sector. Government decision approvals should expire when the statutory or policy basis changes. A decision to process data under a specific legal basis must be reapproved when that legal basis is amended or repealed.
Basic Implementation — Every approval has a defined maximum validity period. Expired approvals are blocked at the infrastructure layer. Expiry metadata is recorded as part of the approval artefact. Advance expiry notifications are sent. This meets minimum mandatory requirements but material-change triggers are not implemented, and renewal is a simple re-confirmation rather than a fresh assessment.
Intermediate Implementation — Material-change triggers are defined for each approval type and linked to authoritative data feeds. Trigger events automatically invalidate affected approvals. Renewal requires a fresh context package with current conditions. Graduated expiry periods are implemented by tier. The system tracks renewal rates, trigger-based invalidation frequency, and the time gap between trigger events and reapproval.
Advanced Implementation — All intermediate capabilities plus: predictive expiry analysis identifies approvals likely to be affected by upcoming events (e.g., scheduled credit reviews, regulatory consultations). Automated analysis detects approvals that are routinely renewed without condition changes, suggesting the approval may be over-conservative or under-reviewed. Independent adversarial testing confirms that expired approval bypass, trigger suppression, and stale renewal attacks all fail.
Required artefacts:
Retention requirements:
Access requirements:
Test 8.1: Time-Based Expiry Enforcement
Test 8.2: Material-Change Trigger Enforcement
Test 8.3: Structural Enforcement (Not Advisory)
Test 8.4: Renewal Requires Fresh Context
Test 8.5: Graduated Expiry by Tier
Test 8.6: Advance Expiry Notification
| Regulation | Provision | Relationship Type |
|---|---|---|
| EU AI Act | Article 9 (Risk Management System) | Supports compliance |
| EU AI Act | Article 14 (Human Oversight) | Direct requirement |
| SOX | Section 404 (Internal Controls Over Financial Reporting) | Supports compliance |
| FCA SYSC | 6.1.1R (Systems and Controls) | Supports compliance |
| NIST AI RMF | MANAGE 2.2 (Risk Mitigation) | Supports compliance |
| ISO 42001 | Clause 9.1 (Monitoring, Measurement, Analysis) | Supports compliance |
| UK GDPR | Article 5(1)(e) (Storage Limitation) | Supports compliance |
Human oversight is only meaningful if it is exercised with reasonable frequency relative to the pace of change. An approval that persists indefinitely without renewal means that human oversight was exercised once and never revisited. Article 14 requires ongoing oversight, which implies periodic reassessment — directly implemented by approval expiry and renewal.
SOX requires periodic assessment of internal controls. An approval that never expires never triggers reassessment, creating a gap in the control evaluation cycle. Approval expiry ensures that financial controls are periodically revalidated, which SOX auditors expect.
While primarily addressing data retention, the storage limitation principle reflects the broader concept that decisions based on data or conditions should be reviewed when those conditions change. Approval expiry applies this principle to governance decisions — ensuring that approvals remain valid only while the conditions supporting them persist.
| Field | Value |
|---|---|
| Severity Rating | High |
| Blast Radius | Organisation-wide — stale approvals enable agents to operate under outdated conditions across multiple business functions |
Consequence chain: Without approval expiry, agents continue operating under approvals that reflect conditions that no longer exist. The longer the approval persists without renewal, the greater the drift between the approved conditions and actual conditions. When the drift produces an adverse outcome, the organisation discovers that the approval predated the condition change. The regulatory consequence is that the organisation cannot demonstrate current, valid approval for the agent's actions — the approval that existed is revealed as stale. The financial consequence depends on the nature of the condition change: a credit downgrade may result in counterparty losses, a regulatory change may result in unlawful processing, and a strategic shift may result in misaligned expenditure. The reputational consequence includes loss of confidence in the organisation's governance lifecycle management.
Cross-references: AG-010 (Time-Bounded Authority Enforcement) addresses time-based authority constraints that complement approval expiry. AG-289 (Task-Scoped Authority Binding Governance) defines the scope that an approval authorises. AG-292 (Approval Context Completeness Governance) defines the context required for both initial approval and renewal. AG-294 (Delegation Revocation Propagation Governance) addresses the propagation of revocation when an approval is invalidated. AG-298 (Post-Approval Mutation Detection Governance) detects changes to the approved action after approval, which may trigger the need for reapproval. AG-170 (Approval Quality and Substantive Review) ensures renewal decisions are substantive. Siblings in this landscape: AG-289 through AG-298.