Governing Data Integrity Risks in Digital Construction: A Qualitative Liability Framework for Generative AI Across the AEC Lifecycle and Smart City Operations

Abstract

Generative artificial intelligence (GenAI) can accelerate Building Information Modeling (BIM) and digital-twin workflows, but probabilistic outputs and manipulated inputs create data-integrity risks that conventional construction-software liability rules do not fully explain. This study develops a reproducible doctrinal and comparative-law method that maps the GenAI value chain onto the Architecture, Engineering, and Construction (AEC) lifecycle. Its principal advantage over single-actor product or professional-negligence analyses is a qualitative proportional-liability assessment framework that evaluates technical control, foreseeability, verification capacity, causal contribution, and evidence preservation across developers, deployers, and professional users. The analysis identifies the following two recurrent pathways of harm: autonomous hallucination entering safety-relevant design information and adversarial or erroneous data entering digital-twin feedback loops. It also specifies syntax, semantic, cybersecurity, and human-verification controls linked to AEC information-management and quality standards. The framework is not an empirically calibrated formula and does not determine legally binding percentages. Instead, it provides a transparent decision aid for applying existing tort doctrines and regulatory duties to distributed GenAI-enabled construction workflows.