This study explores a conceptual framework for integrating Artificial Intelligence (AI) into Enterprise Resource Planning (ERP) systems, emphasising its transformative potential in highly automated industrial environments, often referred to as ‘dark factories’, where operations are carried out with minimal human intervention using robotics, AI, and IoT. These lights-out manufacturing environments demand intelligent, autonomous systems that go beyond traditional ERP functionalities to deliver sustainable enterprise operations and supply chain management. Drawing from secondary data and a comprehensive review of existing literature, the study identifies significant gaps in current AI-ERP research and practice, namely, the absence of a unified adoption framework, limited focus on AI-specific implementation challenges, and a lack of structured post-adoption evaluation metrics. In response, this paper proposes a novel integrated conceptual framework that combines the Technology–Organisation–Environment (TOE) framework, the Technology Acceptance Model (TAM), and the Information Systems (IS) Success Model. The model incorporates industry-specific dark factors, such as AI autonomy, human–machine collaboration, operational agility, and sustainability, by optimising resource efficiency, enabling predictive maintenance, enhancing supply chain resilience, and supporting circular economy practices. The primary research aim of the current study is to provide a theoretical foundation for further empirical research on the input of AI-ERP systems into autonomous industry settings. The framework provides a robust theoretical foundation and actionable guidance for researchers, technology leaders, and policy-makers navigating the integration of AI and ERP in sustainable enterprise operations and supply chain management.

Brain tumour classifiers are often reported as near-perfect and explainable without checking patient structure or annotations. We reproduced a recent attention framework (SSPANet) on the Figshare dataset (3064 T1-weighted contrast-enhanced slices, 233 patients). Its image-level split is fully patient-leaky; every configuration exceeds 98% accuracy, which drops by 5–9 points on the patient-wise split. Under a fixed training recipe, seven attention modules stayed within ±0.63 points of the plain backbone, whereas the recipe changed accuracy by up to 4.7 points. As a case study, TG-SSPANet uses expert masks to supervise attention, train a segmentation decoder and gate evidence pooling. With ResNet50, it reached 96.8% slice-level and, averaging slice probabilities, 98.3% patient-level accuracy (84.5% if every slice had to be correct; 15% of patients had discordant slices), within the confidence intervals of simpler mask-using pipelines; its attention module added no measurable benefit. Its Grad-CAM++ maps hit the tumour in 81% of held-out slices, but this follows from its decoder by construction, not the classifier’s reasoning. After tumour inpainting, whole-slice models kept 80–86% accuracy, and shuffling the remaining tissue left 60–74%, so location, perilesional cues and tissue statistics cannot be separated. Patient-wise evaluation, same-recipe controls and mask-based faithfulness tests should precede explainability claims.