AI can obey every limit in a company’s delegation of authority while making thousands of independent, individually permitted decisions under the same authority, with no separate limit on how often that authority is exercised. The written delegation can remain unchanged while the amount of corporate action passing through it increases substantially.
Delegation frameworks usually express authority one decision at a time. A manager may approve a discount within a defined range. Counsel may accept an approved contractual position without further escalation. Where those powers are assigned to human roles, substantial growth in the number of decisions generally requires additional organisational capacity through staffing or a reallocation of work.
The relevant change occurs when an autonomous system inherits that existing authority structure. Deloitte’s 2026 enterprise interviews describe uses of AI agents in financial services, aviation and manufacturing. Existing authority matrices can fit naturally into that environment because they already define the boundary for an individual decision and the point at which further approval is required.
Traditional aggregation controls address related risks. Budgets can constrain cumulative spending and aggregation provisions can combine transactions that form part of the same arrangement. The issue here concerns genuinely independent decisions whose wider economic or legal effect develops across repeated exercise.
A discount authority illustrates the distinction. Each customer decision may be unrelated to the next and every discount may remain within the approved percentage. An autonomous system can exercise that authority far more extensively within the same function, while the accumulated effect becomes visible later in margin or pricing patterns.
Contractual fallbacks can produce a similar result. An approved liability position or warranty concession may be acceptable in an individual negotiation and require no escalation. Repeated autonomous use across independent contracts can alter the company’s exposure while every agreement remains within the approved playbook.
Boards should address this before autonomous authority is deployed. They should require management to identify existing powers that an AI system may exercise autonomously where each independent decision is constrained and repeated independent exercise is not separately bounded.
For every authority that meets that test, boards should require an authority velocity limit before autonomous exercise begins. The limit determines how much independent exercise of the same delegated authority may occur within a defined period before further approval or escalation is required.
The period is integral to the control. Increasing human decision capacity normally takes organisational time. Autonomous execution can compress a comparable volume of decisions into a much shorter interval. A thousand authorised decisions reached over a year and the same number reached in a day can present different governance consequences even when the boundary applied to every individual decision is identical.
Management should select a measure that reflects the authority involved. Decision count may suit repeated waivers or approvals. Aggregate exposure may better capture contractual or financial commitments. The board’s role is to require the additional limit and receive enough information to assess whether it remains appropriate.
Responsibility extends across the accumulated effect of authorised decisions. The business function that owns the delegated power remains accountable for the consequences produced through autonomous exercise. It should calibrate the authority velocity limit to the underlying risk and maintain an escalation path appropriate to that authority.
Boards should also require reporting against those limits. Reporting should show the extent of autonomous exercise and proximity to the applicable limit, with triggered escalations identified separately. It should give the board enough information to assess whether repeated exercise is producing consequences that justify recalibration or a different approval path.
For companies reporting against the UK Corporate Governance Code 2024, the issue also has an internal-control dimension. Provision 29 says that the board should monitor the company’s risk management and internal control framework and, at least annually, carry out a review of its effectiveness. The monitoring and review should cover all material controls. For financial years beginning on or after 1 January 2026, the annual report should also include a declaration on the effectiveness of material controls at the balance sheet date.
The FRC Corporate Governance Code Guidance leaves the identification of material controls to each board. Paragraph 272 gives examples that include controls over information and technology risks, including new technologies such as artificial intelligence. Paragraph 297 adds an important point about effectiveness. When establishing whether a material control is operating effectively, the board should also consider its effective design and implementation.
That design question fits delegated authority closely. A control may operate exactly as designed at the level of every individual decision while containing no separate boundary for repeated autonomous exercise of the same power. Where the board has identified delegated authority as a material control, it should assess whether that design remains effective for the way the authority is now being exercised.
Transaction-level compliance still establishes whether individual acts remained within their authorised boundaries. Authority velocity reporting shows how extensively the same authority is being exercised across independent decisions and how close that exercise is to the point at which further approval is required.
The governance problem begins when every autonomous decision is authorised and the delegation is still doing exactly what it was designed to do.