Loading...
Loading...
The level of dependence on AI system outputs beyond which institutional functioning becomes degraded if the system is removed or fails. COA specifies reliance thresholds and requires governance review when they are approached, preventing unintended irreversibility.
A reliance threshold might specify: 'If more than 70% of decisions in a category follow model recommendations without substantive human modification for six consecutive months, a delegation review is triggered.' This catches authority drift before it becomes irreversible.
Section 5.1: Delegation Boundaries