Execution Drift refers to a change in AI behavior caused by shifts in execution context rather than by model updates. These shifts may come from changes in data state, prompt context, or pipeline structure.