RISKO-RIENTED ADMISSION OF AI SOLUTIONS TO AUTOMATIC EXECUTION OF TASKS IN BPM SYSTEMS BASED ON CRITICALITY POLICIES
Abstract
The paper addresses the admission of AI decisions to automated execution of tasks in BPM systems with bot executors after the process selection stage. The focus is on formalizing and validating a safe transition from AI risk assessment to real action in an executable process environment. A riskoriented method is proposed and tested in which the admissibility of automatic execution is determined with respect to the process and the task class, with a ban on risk deescalation, mandatory executionpolicy checks, confidence control, and verification of bot availability. The results show that the practical value of intelligent automation depends not only on process selection quality but also on the ability to safely carry AI decisions through to execution with a controlled handoff to manual handling.