Authors : Anil Kumar Pakina

Volume/Issue : Volume 10 - 2025, Issue 12 - December

Google Scholar : https://tinyurl.com/3r4mhb9s

Scribd : https://tinyurl.com/2v9j63hh

DOI : https://doi.org/10.38124/ijisrt/25dec1364

Note : A published paper may take 4-5 working days from the publication date to appear in PlumX Metrics, Semantic Scholar, and ResearchGate.

Abstract : The recent developments of artificial intelligence technologies, which shift towards more autonomy and adaptive learning, have demonstrated the inherent weaknesses in the existing explainability and regulatory systems. The paradigms of conventional explainable AI are mostly post-hoc and relatively static and rely on the assumption that the behavior of the models will be the same once deployed. However, autonomous AI systems do not stand at a single place of operation but are constantly evolving thus making single-case explanations ineffective when it comes to ensuring the long-term accountability, security, and compliance with regulatory standards. The current paper thus suggests Continuous Explainability Auditing (CEA) as an alternative to governance, while redefining explainability as an audit service (iterative, run-time, rather than retrospective) of an interpretive artifact. CEA enables the acquisition and analysis of decision rationales and behavioral patterns and evidence of policy compliance in autonomous AI systems operating in dynamic and high-risk environments. The framework can detect behavioral drift, misalignment, regulatory deviation and adversarial manipulation by embedding explainability outputs into a governance control loop, which uses risk thresholds and compliance triggers to detect the presence of all unwanted behaviors at their initial stages. Compared to traditional explainability systems, CEA puts more emphasis on temporal traceability and longitudinal reasoning analysis, both of which maintain the performance of the system and meet privacy limitations through distributed, minimally invasive monitoring systems. The practical feasibility of the suggested paradigm is explained by the case studies of regulated financial and cyber- security areas, where autonomous AI agents are prone to the strict transparency and auditability requirements. The findings reveal that CEA allows proactive control, enables evidence that is ready to be provided to regulators and allows governance at the scale of operational workflows without negatively influencing workflows. Together, this demonstrates the fact that ongoing explainability is an essential rather than a supplementary governance requirement of safe, reliable and compliant deployment of autonomous AI systems in regulated industries.

Keywords : Continuous Explainability Auditing; Explainable AI (XAI); AI Governance; Autonomous AI Systems; Regulatory Compliance; Runtime Auditing; Algorithmic Accountability; Trustworthy AI.

References :

1. Abebe, T., & Müller, S. (2024). Towards autonomous audit pipelines in AI governance systems: A transparency engineering review. Journal of Artificial Intelligence Policy, 6(2), 45–66.
2. Ahmed, R., & Lin, Z. (2024). Runtime drift detection in evolving neural architectures. International Journal of Machine Systems, 18(1), 77–95.
3. Alonso, J. (2023). Adaptive model accountability and the future of algorithmic risk structures. AI Governance Review, 2(4), 101–118.
4. Aoyon, R. S., & Hossain, I. (2023, December). A chatbot based auto-improving health care assistant using RoBERTa. In 2023 3rd International Conference on Robotics, Automation and Artificial Intelligence (RAAI) (pp. 213-217). IEEE.
5. Gad, E., Abou Khatwa, M., A. Elattar, M., & Selim, S. (2023, July). A novel approach to breast cancer segmentation using U-Net model with attention mechanisms and FedProx. In Annual Conference on Medical Image Understanding and Analysis (pp. 310-324). Cham: Springer Nature Switzerland.
6. Banerjee, K., & Cho, H. (2025). Continuous monitoring mechanisms for adversarial indicators in medical transformers. Neural Systems and Security, 33(2), 201–220.
7. Bennett, R., & Alvarez, D. (2024). Federated transparency and distributed learning challenges. Journal of Digital System Integrity, 5(3), 90–112.
8. Chen, P. (2023). Explainability pipelines for healthcare analytics: A lifecycle study. Medical Informatics & Computing, 7(2), 55–74.
9. Desai, V., & Singh, M. (2025). Risk-aware model supervision in autonomous AI environments. International Journal of Emerging Compute, 14(1), 30–51.
10. Ding, L. (2024). Conceptual frameworks for AI audit embedding. Journal of Computational Ethics, 9(3), 156–170.
11. Sadik, R., Rahman, T., Bhattacharjee, A., Halder, B. C., & Hossain, I. (2025). Exploring Adversarial Watermarking in Transformer-Based Models: Transferability and Robustness Against Defense Mechanism for Medical Images. arXiv preprint arXiv:2506.06389.
12. Fang, W., & Zhao, H. (2025). Multi-domain auditing adaptation for deep learning supervision. Autonomous Intelligence Studies, 12(1), 41–62.
13. George, K., & Patel, H. (2024). Algorithmic integrity in transformer-based clinical applications. Clinical Decision Informatics, 8(2), 66–89.
14. Haddad, N. (2025). Continuous accountability loops in cognitive robotics. Robotics and Control Systems, 17(2), 113–140.
15. Hussain, T., & Malik, S. (2024). Runtime explainability for human-AI safety models. Human-Centered AI Journal, 3(1), 22–35.
16. Ibrahim, O. (2023). The traceability problem in autonomous frameworks. Journal of AI Modeling, 11(4), 72–94.
17. Aoyon, R., Hossain, I., Abdullah-Al-Wadud, M., & Uddin, J. (2025). A secured and continuously developing methodology for breast cancer image segmentation via u-net based architecture and distributed data training. Computer Modeling in Engineering & Sciences, 142(3), 2617.
18. Jiang, X. (2024). Deep model governance and privacy-adapted auditing. Security & Computing, 20(3), 177–203.
19. Kim, J., & Lee, J. (2024). Quantifying explainability decay rates in evolving networks. Neural Transparency Research, 5(1), 130–148.
20. Kumar, R. (2025). A review of post-deployment reconfiguration in medical AI analytics. Journal of Applied Clinical AI, 9(2), 211–235.
21. Aoyon, R. S., & Hossain, I. (2024, February). A novel approach of making French language learning platform via brain-computer interface and deep learning. In International Congress on Information and Communication Technology (pp. 399-409). Singapore: Springer Nature Singapore
22. Li, Q., & Zhang, M. (2023). Self-learning AI and ethics in continuous decision models. AI & Law Review, 6(3), 109–122.
23. Liu, S., & Wong, C. (2024). Distributed reasoning analysis in multi-node segmentation systems. Scientific Computing Letters, 13(4), 88–106.
24. Martinez, A. (2024). Audit-centric transparency architectures for high-risk neural systems. AI Ethics Quarterly, 10(2), 33–54.
25. Nguyen, P., & Carter, J. (2024). Runtime interpretation in transformer training workflows. Deep Intelligence Studies, 4(1), 149–162.
26. Olsen, M. (2025). Lifelong explainability architectures: A conceptual mapping. Neural Policy Journal, 3(1), 19–44.
27. Rahman, K., & Omar, A. (2024). Model drift and interpretability decay in U-Net systems. Healthcare AI Studies, 21(2), 55–70.
28. Singh, V., & Rao, R. (2025). Real-time monitoring logic for continuous model integrity. Journal of Advanced Computing Theory, 18(2), 92–115.
29. Torres, B. (2024). Regulator-ready neural auditing structures: A policy review. AI Regulatory Perspectives, 8(3), 50–78.
30. Yuan, J. (2023). Adversarial evolution testing in autonomous systems. Security Engineering & AI, 7(1), 99–121.