Abstract
The developing integration of Artificial Intelligence (AI) within project management has engendered novel decision-making challenges, particularly amid escalating project investments. Despite extensive discourse on Escalation of Commitment, there is a dearth of investigations into the dynamics of De-escalation of Commitment (DoC), especially when AI-based control analyses are employed as early-warning mechanisms. This research employed semi-structured interviews with 22 project managers spanning diverse industries to explore the interplay between AI control signals and managerial adjudication in project escalation contexts. Drawing upon this data, a hitherto unarticulated behavioural construct, termed override AI bias is delineated. The construct encapsulates the propensity of managers to discount AI alerts, a phenomenon modulated by variables such as AI explainability, project uniqueness and information asymmetry. These findings offer critical insights into the tensions at the human-algorithm interface, with implications for both theoretical understanding and practical implementation of AI-augmented project control strategies.