An AI initiative can remain on track after its approved purpose has changed. A system introduced for internal work starts shaping customer-facing recommendations, while the board pack still reports delivery as planned. Was the wider use authorized before it began?
Why It Matters
Delegation has a boundary
Management selects, deploys and operates AI systems. A management steering group may coordinate assurance and risk reviews. A board committee may scrutinize matters within its terms of reference. The full board exercises the strategic and oversight responsibilities assigned to it under applicable law and the company's governance arrangements. These functions intersect, but the authority attached to each is different.
The G20/OECD Principles of Corporate Governance describe board committees as supporting the full board, which ordinarily retains responsibility for committee decisions unless law provides otherwise [1]. NIST's voluntary AI Risk Management Framework addresses documented AI risk roles and communication lines (GOVERN 2.1), and executive leadership's responsibility for AI risk decisions (GOVERN 2.3) [2]. These sources do not require directors to approve individual AI releases or prescribe a universal committee model.
A delegated approval is useful because it lets management act within defined limits. It does not automatically extend to a materially different purpose, affected group or degree of autonomy.
When the permitted use changes
Consider a hypothetical company that authorizes an AI assistant to prepare internal product briefings. The original approval limits its use to employees. Six months later, management proposes allowing the assistant to generate recommendations that customers will see in several markets. Technical and risk reviews are completed, and the delivery report remains positive.
The wider use may change customer exposure and applicable obligations. It could remain within management's authority, require notification to a board committee or cross a matter reserved for full-board approval. That depends on documented decision limits and applicable requirements, not the fact that the same AI system is being used.
Approval for one use is not automatic approval for another. Someone must check the new proposal against the earlier authorization before the new scope is activated.
6xD Interpretation
In the 6xD view of the Digital Cognitive Organization (D2), intelligence is embedded in enterprise workflows [3]. When its permitted use changes, the relevant decision authority may change with it. This is a 6xD editorial interpretation, not a new legal requirement.
Executive Implications
What happens when approval is uncertain?
For a material change that may exceed the original mandate, management should use the company's existing referral process rather than assume approval. The responsible executive should identify who may defer or restrict the expansion while the appropriate authority is clarified. The response depends on the company's risk appetite, reserved matters and operating arrangements; it is not a universal stop rule.
Directors need visibility of material boundary crossings, late approvals and unresolved referrals, with the decision and corrective action recorded. They do not need every model test or minor change request. This is the distinction between retaining oversight and taking over execution.
Recommended Actions
The board chair's two-record test
For the next material expansion of an AI use, ask the accountable executive, with the company secretary or relevant governance lead, to produce two dated records.
The first is the original authorization: what use was permitted, which limits applied and who held the authority to approve changes. The second is the changed-scope decision: what was proposed, which threshold was checked, who decided or received the referral, and whether the decision preceded activation.
Ask one question: Can management demonstrate that the expanded use was within the authority granted when it went live?
If the second record shows approval after activation, directors should ask how the exception was handled and whether agreed limits operated as intended. If no suitable change has occurred, test one plausible expansion against the actual delegation. Neither approach requires the board to manage routine releases.
Sources
- 01OECD (2023), G20/OECD Principles of Corporate Governance 2023, V.D and V.E.2 (2023)
- 02 NIST (2023), AI Risk Management Framework 1.0, GOVERN 2.1 and 2.3 (2023)
- 03Niango, S. (2026), The Six Digital Perspectives, Vol. 00, D2. DigitalQatalyst
- 046xD| Insight article: Why AI Governance Is Becoming a Board-Level Responsibility



