The growing use of agentic artificial intelligence in project management has created a profound responsibility gap, yet prevailing legal doctrines maintain that artificial intelligence itself bears no legal liability. This study aims to systematically review the emerging literature to examine how liability should be governed when agentic artificial intelligence shapes project decisions in the built environment.
A systematic review, following the Preferred Reporting Items for Systematic Reviews and Meta Analyses 2020 guidelines, synthesised 16 peer reviewed studies published between 2016 and 2025. The review focused on governance models, liability doctrines and practical mechanisms for allocating responsibility across the legal, ethical and technical dimensions of artificial intelligence-enabled project management.
Across all included studies, shared accountability emerged as the dominant and conceptually robust governance framework, typically operationalised through human-in-the-loop oversight and embedded within legal, ethical and technical controls. Strict liability for any single actor and models that attribute responsibility exclusively to either humans or artificial intelligence were consistently rejected. However, the review exposes major gaps: there are no mature, phase-specific processes for embedding shared responsibility throughout the project lifecycle; no clear legal standards for resolving conflicts between opaque black-box outputs and human judgments; and fragmented, cross-jurisdictional regulations that hinder consistent practice.
This study is limited by its restricted data set, focus on built environment literature and the still-evolving nature of research on liability in agentic artificial intelligence, which may have excluded some relevant studies from other databases, disciplines and grey literature sources.
The findings of this study indicate that organisations should move beyond high-level principles of shared accountability and develop detailed protocols for responsibility allocation at each project phase, including documentation, escalation and dispute-resolution mechanisms when human and artificial-intelligence recommendations diverge. Regulators and professional bodies need to translate emerging jurisprudential insights into enforceable standards and guidance, particularly for cross border projects that rely on autonomous decision support systems.
This review integrates jurisprudential, governance and technical perspectives on agentic artificial intelligence in project management, positioning shared accountability with human-in-the-loop oversight as the most advanced, yet still under-operationalised, framework. It reframes the responsibility gap not as an abstract ethical dilemma but as a concrete legal and procedural challenge, arguing that only phase specific protocols and harmonised standards can sustain legitimate, trustworthy use of increasingly autonomous artificial intelligence in project management.
