Trustworthy Edge Generative AI: A Safety and Robustness Evaluation Framework for On-Device Image Synthesis Models
Keywords:
Edge generative AI, on-device image synthesis, trustworthiness, safety, robustness, evaluation framework, socio-technical systems, adversarial robustness, fairness, edge infrastructureAbstract
The rapid proliferation of generative artificial intelligence has spurred immense interest in deploying image synthesis models directly on edge devices, ranging from smartphones to embedded IoT platforms. This migration promises reduced latency, enhanced privacy, and operation in connectivity-constrained environments. However, the confluence of generative capabilities with the resource and security constraints of edge hardware introduces profound trustworthiness challenges that remain underexamined from a systems perspective. This paper presents a comprehensive evaluation framework for the safety and robustness of on-device image synthesis models, explicitly addressing adversarial resilience, fairness across demographic representations, data leakage risks, operational stability under varying hardware conditions, and compliance with evolving regulatory mandates. The framework is constructed upon an interdisciplinary foundation that integrates systems architecture, socio-technical governance, and sustainability considerations. The discussion systematically analyzes the structural trade-offs inherent in compressing large foundation models for edge deployment, the fragility induced by aggressive quantization and pruning, and the systemic risks of deploying generative models within heterogeneous hardware ecosystems. By focusing on system-level interactions rather than isolated algorithmic assessment, the paper delineates how edge-specific factors, such as intermittent connectivity, energy budgets, and federated learning pipelines, reshape the threat surface. Policy implications are examined through the lens of the European Union Artificial Intelligence Act and emerging technical standards, revealing critical gaps between the pace of edge generative AI innovation and the maturity of assurance mechanisms. The evaluation framework is proposed as a socio-technical instrument that can guide developers, regulators, and infrastructure providers toward a trustworthy edge AI ecosystem, balancing performance imperatives with societal and ethical obligations.
References
1. Ho, J., Jain, A., & Abbeel, P. (2020). Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems (NeurIPS).
2. Rombach, R., Blattmann, A., Lorenz, D., Esser, P., & Ommer, B. (2022). High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
3. Li, Y., Yao, Z., Zhang, L., Wang, S., Li, S., & Zhang, X. (2024). MobileDiffusion: Subsecond text-to-image generation on mobile devices. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).
4. Ignatov, A., Timofte, R., Kulik, A., Yang, S., Wang, K., Baum, F., ... & Van der Laan, B. (2019). AI benchmark: All about deep learning on smartphones in 2019. In IEEE/CVF International Conference on Computer Vision Workshops (ICCVW).
5. 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).
6. Chen, T., Moreau, T., Jiang, Z., Zheng, L., Yan, E., Shen, H., ... & Krishnamurthy, A. (2018). TVM: An automated end-to-end optimizing compiler for deep learning. In Proceedings of the USENIX Symposium on Operating Systems Design and Implementation (OSDI).
7. McMahan, B., Moore, E., Ramage, D., Hampson, S., & y Arcas, B. A. (2017). Communication-efficient learning of deep networks from decentralized data. In Proceedings of the International Conference on Artificial Intelligence and Statistics (AISTATS).
8. Konečný, J., McMahan, H. B., Yu, F. X., Richtárik, P., Suresh, A. T., & Bacon, D. (2016). Federated learning: Strategies for improving communication efficiency. In NIPS Workshop on Private Multi-Party Machine Learning.
9. Strubell, E., Ganesh, A., & McCallum, A. (2019). Energy and policy considerations for deep learning in NLP. In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL).
10. Floridi, L., Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., ... & Schafer, B. (2018). AI4People—An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations. Minds and Machines, 28(4), 689-707.
11. European Commission. (2021). Proposal for a Regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). COM(2021) 206 final.
12. Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. In Proceedings of the Conference on Fairness, Accountability and Transparency (FAT).
13. Chen, C., Wang, C., Li, Y., Wan, Z., Geng, M., Xiao, J., ... & Peng, Y. (2026). JuZhou 1.0 Technical Report: The First Edge-Native Text-to-Image Foundation Model Trained Entirely on China-Developed AI Accelerators. arXiv preprint arXiv:2606.28421.
14. Binns, R. (2018). Fairness in machine learning: Lessons from political philosophy. In Proceedings of the Conference on Fairness, Accountability and Transparency (FAT).
15. Fredrikson, M., Jha, S., & Ristenpart, T. (2015). Model inversion attacks that exploit confidence information and basic countermeasures. In Proceedings of the ACM Conference on Computer and Communications Security (CCS).
16. Song, C., Ristenpart, T., & Shmatikov, V. (2017). Machine learning models that remember too much. In Proceedings of the ACM Conference on Computer and Communications Security (CCS).
17. Carlini, N., Tramèr, F., Wallace, E., Jagielski, M., Herbert-Voss, A., Lee, K., ... & Papernot, N. (2021). Extracting training data from large language models. In Proceedings of the USENIX Security Symposium.
18. Madry, A., Makelov, A., Schmidt, L., Tsipras, D., & Vladu, A. (2018). Towards deep learning models resistant to adversarial attacks. In International Conference on Learning Representations (ICLR).
19. Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). "Why should I trust you?" Explaining the predictions of any classifier. In Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD).
20. Reddi, V. J., Cheng, C., Kanter, D., Mattson, P., Schmuelling, G., Wu, C. J., ... & Zlateski, A. (2020). MLPerf inference benchmark. In Proceedings of the International Symposium on Computer Architecture (ISCA).
21. Patterson, D., Gonzalez, J., Le, Q., Liang, C., Munguia, L., Rothchild, D., ... & Dean, J. (2021). Carbon emissions and large neural network training. arXiv preprint arXiv:2104.10350.
22. National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST AI 100-1.
23. Wu, C. J., Brooks, D., Chen, K., Chen, D., Choudhury, S., Dukhan, M., ... & Zhang, Y. (2022). Sustainable AI: Environmental implications, challenges and opportunities. In Proceedings of Machine Learning and Systems (MLSys).
24. Choi, K., Grover, A., Singh, T., Shu, R., & Ermon, S. (2020). Fair generative modeling via weak supervision. In International Conference on Machine Learning (ICML).
25. Gebru, T., Morgenstern, J., Vecchione, B., Vaughan, J. W., Wallach, H., Daumé III, H., & Crawford, K. (2021). Datasheets for datasets. Communications of the ACM, 64(12), 86-92.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Data Intelligence and AI Systems
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.
