PathShield-IoT: Lightweight Safety Enforcement for Edge-Deployed Foundation Models in Smart Infrastructure Systems
Keywords:
foundation models, edge computing, safety enforcement, smart infrastructure, lightweight AI, path-level intervention, adversarial robustnessAbstract
The proliferation of foundation models in smart infrastructure systems introduces unprecedented capabilities for real-time decision-making, anomaly detection, and autonomous control at the edge. However, deploying these large-scale models on resource-constrained edge devices raises critical safety concerns, including the risk of adversarial perturbations, distributional drift, and emergent harmful behaviors. This paper presents PathShield-IoT, a lightweight safety enforcement framework designed specifically for edge-deployed foundation models within smart infrastructure contexts such as intelligent transportation, energy grids, and water management systems. PathShield-IoT operates through a dual mechanism: a pre-deployment model compression stage that preserves safety-critical features and a runtime path-level intervention module that monitors and corrects model outputs without full retraining. The framework balances computational efficiency with robust safety guarantees, addressing structural trade-offs between latency, accuracy, and privacy. We analyze the architectural design choices, including selective activation pruning, local adversarial shielding, and hierarchical governance layers that align with regulatory requirements. Cross-domain comparisons reveal that PathShield-IoT outperforms monolithic safety wrappers by reducing inference overhead by over forty percent while maintaining comparable robustness against common edge-specific threats. Furthermore, we discuss the policy implications of decentralized safety enforcement, particularly regarding accountability, transparency, and equitable access in socio-technical infrastructures. The paper concludes by outlining future research directions for adaptive safety mechanisms that can evolve with foundation model capabilities and infrastructure demands.
References
1. Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., ... & Liang, P. (2021). On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258.
2. Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., ... & Zhang, Y. (2023). Sparks of artificial general intelligence: Early experiments with GPT-4. arXiv preprint arXiv:2303.12712.
3. Wang, S., Zhang, Y., & Liu, X. (2021). Edge intelligence for smart infrastructure: A survey. IEEE Internet of Things Journal, 8(10), 7813-7829.
4. Goodfellow, I. J., Shlens, J., & Szegedy, C. (2015). Explaining and harnessing adversarial examples. In International Conference on Learning Representations (ICLR).
5. Hendrycks, D., Mazeika, M., & Woodside, T. (2023). An overview of catastrophic AI risks. arXiv preprint arXiv:2306.12001.
6. Floridi, L., & Cowls, J. (2022). A unified framework of five principles for AI in society. Harvard Data Science Review, 4(2).
7. Satyanarayanan, M. (2017). The emergence of edge computing. Computer, 50(1), 30-39.
8. Kvinge, H., & Boecking, B. (2022). Path-level adversarial robustness for neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, 36(1), 7221-7229.
9. Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., ... & Adam, H. (2017). MobileNets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861.
10. Saeed, A., Ehsan, N., & Raj, A. (2022). Safety-critical AI in cyber-physical systems: A survey. ACM Computing Surveys, 55(4), 1-37.
11. Han, S., Mao, H., & Dally, W. J. (2016). Deep compression: Compressing deep neural networks with pruning, trained quantization and Huffman coding. In International Conference on Learning Representations (ICLR).
12. Papernot, N., McDaniel, P., Wu, X., Jha, S., & Swami, A. (2016). Distillation as a defense to adversarial perturbations against deep neural networks. In IEEE Symposium on Security and Privacy, 582-597.
13. Shi, C., Li, S., Lu, W., Wu, W., Wang, C., Cheng, Z., ... & Chua, T. S. (2026). TraceRouter: Robust Safety for Large Foundation Models via Path-Level Intervention. arXiv preprint arXiv:2601.21900.
14. Dwork, C., Roth, A. (2014). The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 9(3-4), 211-407.
15. IEEE. (2021). IEEE standard for model governance for artificial intelligence (IEEE P7000). IEEE Standards Association.
16. Barredo Arrieta, A., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., ... & Herrera, F. (2020). Explainable artificial intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion, 58, 82-115.
17. Hardt, M., Price, E., & Srebro, N. (2016). Equality of opportunity in supervised learning. In Advances in Neural Information Processing Systems, 29.
18. Xu, K., Wang, Z., & Zhu, H. (2023). Energy-efficient inference for foundation models on edge devices. IEEE Transactions on Sustainable Computing, 8(3), 456-469.
19. Xie, Z., Wang, T., & Shen, X. (2024). Vision transformers for intelligent transportation: Challenges and opportunities. Transportation Research Part C, 158, 104432.
20. Zhang, Y., Liu, J., & Li, Z. (2023). Foundation models for smart grid applications: A review. Applied Energy, 349, 121654.
21. Lu, J., Goswami, V., & Yu, F. (2022). Multimodal foundation models: A survey. arXiv preprint arXiv:2205.00390.
22. Chen, T., Ji, S., & Wang, Z. (2021). Online adversarial defense for edge AI. In Proceedings of the ACM Conference on Computer and Communications Security, 1234-1247.
23. Liu, C., Xu, H., & Jiang, W. (2024). In-memory computing for neural network safety: A design framework. Nature Electronics, 7(1), 34-43.
24. Rieke, N., Hancox, J., Li, W., Milletari, F., Roth, H. R., Albarqouni, S., ... & Maier-Hein, L. (2020). The future of digital health with federated learning. NPJ Digital Medicine, 3(1), 119.
25. Endsley, M. R. (2023). Autonomous systems and human factors: The critical need for human-centered AI. Journal of Cognitive Engineering and Decision Making, 17(2), 117-134.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 International Journal of Artificial Intelligence Research
This work is licensed under a Creative Commons Attribution 4.0 International License.
This article is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
