AL-HARIS: Advanced Layer for Hacked-Aware Resilient Intelligent Sentinel

Abstract Artificial Intelligence (AI) has become integral to modern drilling operations, supporting applications such as Rate of Penetration (ROP) optimization, early warning systems for drilling trouble prediction, and real-time decision-making. However, the effectiveness of these AI-driven systems depends entirely on the integrity of streaming data from rig-side sensors—data that is vulnerable to tampering, sensor spoofing, and data poisoning attacks. This paper introduces AL- HARIS (Advanced Layer Hacked-Aware Resilient Intelligent Sentinel), a cybersecurity-aware AI framework designed to detect data poisoning attacks in streaming drilling data. Unlike traditional anomaly detection systems that focus on iden- tifying unusual but legitimate operational patterns, AL-HARIS specifically targets malicious data manipulation through a specialized Bidirectional Gated Recurrent Unit (BiGRU) model with conservative labeling strategy. We evaluate AL- HARIS using the public Volve University of Stavanger Rate of Penetration (USROP) dataset with synthetic data poison- ing attacks at contamination levels ranging from 10% to 75%. The framework achieves exceptional performance with Receiver Operating Characteristic – Area Under the Curve (ROC-AUC) scores above 0.98 across all wells in leave-one- well-out cross-validation, demonstrating robust generalization across diverse geological conditions. By embedding data integrity verification directly into the data pipeline, AL-HARIS reframes data integrity as a prerequisite for trustworthy AI in drilling operations, providing a foundation for cybersecurity-aware drilling workflows and ensuring the reliability of AI-driven decision-making in safety-critical environments.