大模型在智慧家庭应用中的安全挑战与应对策略

doi:10.19784/j.cnki.issn1672-0172.2025.02.014

摘要/Abstract

摘要： 以GPT系列为代表的生成式大模型技术的迅猛发展给交互类产品与服务带来了深刻变革。相较于传统模型,大模型带来生成内容不可控、易被攻击、模型黑盒等新的安全问题,从而制约其落地应用。在各类应用场景中,智慧家庭由于关联多种家庭设备且面向多类人群,对安全性要求更高。该研究聚焦上述问题,深入剖析大模型在智慧家庭落地中面临的内容、设备、隐私三大安全问题,并针对性地给出应对之策,旨在助力构建新一代安全智能、个性多元的家庭服务体系。

关键词: 人工智能安全性, 大模型, 大模型应用, 智慧家庭

Abstract: The rapid development of generative LLMs, represented by the GPT series, has profoundly changed intelligent interactive products and services. Compared to traditional models, LLMs have new security issues such as hallucination, vulnerability to attacks, and uninterpretability, which significantly hinders their practical applications. Among various application scenarios, smart home has higher requirements for security due to its association with multiple devices and a diverse range of target users. This paper focuses on this scenario, analyzing the three major security issues of content, devices, and privacy that LLMs face in the application of smart homes. Additionally, the targeted solutions are proposed accordingly, which aims to the establishment of the new generation smart home ecosystem that are secure, intelligent, personalized and diversified.

Key words: AI security, Large language models, LLM application, Smart home

中图分类号:

TM925
TP29

邵研, 陆海俊, 罗红. 大模型在智慧家庭应用中的安全挑战与应对策略[J]. 家电科技, 2025, 0(2): 94-97.

SHAO Yan, LU Haijun, LUO Hong. Security challenges and countermeasures of LLMs’ application to smart home[J]. Journal of Appliance Science & Technology, 2025, 0(2): 94-97.

/ 推荐

参考文献

[1] Alec Radford, Karthik Narasimhan, Tim Salimans, et al.Improving Language Understanding by Generative Pre-training[R]. OpenAI, 2018.
[2] Guangxuan Xiao, Ji Lin, Mickael Seznec, et al.SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models[C]. International Conference on Machine Learning, 2023: 38087-38099.
[3] Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan, et al.Sparks of Artificial General Intelligence: Early Experiments with GPT-4[R]. CoRR abs/2303.12712, 2023.
[4] 李伟, 贾奇伟, 劳春峰, 等. 基于LLM的家电智能交互控制系统研究[J]. 家电科技, 2024(zk): 125-129.
[5] 田云龙, 王统帅, 牛丽. 智能家居领域利用AIGC大模型实现全屋智慧场景自编排交互体验的系统和方法[J]. 家电科技, 2024(zk): 435-439.
[6] 阿里云计算有限公司, 中国信息通信研究院. 大模型安全研究报告(2024)[R], 2024.
[7] 朱军,董胤蓬. 大模型安全与治理探讨[J]. 中国信息安全, 2023(08): 57-60.
[8] Jingwei Yi, Rui Ye, Qisi Chen, et al.On the Vulnerability of Safety Alignment in Open-Access LLMs[C]. Association for Computational Linguistics, 2024: 9236-9260.
[9] Chuan Guo, Alexandre Sablayrolles, Hervé Jégou, et al.Gradient-based Adversarial Attacks Against Text Transformers[C]. Conference on Empirical Methods in Natural Language Processing, 2021: 5747-5757.
[10] Xiaoqun Liu, Jiacheng Liang, Muchao Ye, et al.Robustifying Safety-Aligned Large Language Models through Clean Data Curation[R]. CoRR, abs/2405.19358, 2024.
[11] Yi Luo, Zhenghao Lin, YuHao Zhang, et al. Ensuring Safe and High-Quality Outputs: A Guideline Library Approach for Language Models[C]. North American Chapter of the Association for Computational Linguistics - Human Language Technologies, 2024: 1152-1197.
[12] OpenAI. OpenAI o1 System Card[R], 2024.
[13] Badhan Chandra Das., M. Hadi Amini, Yanzhao Wu. Security and Privacy Challenges of Large Language Models: A Survey[R]. ACM Computing Surveys, 2025: 1-39.
[14] Ehsan Ullah, Anil Parwani, Mirza Mansoor Baig, et al.Challenges and Barriers of Using Large Language Models (LLM) such as ChatGPT for Diagnostic Medicine with a Focus on Digital Pathology - a Recent Scoping Review[J]. Diagnostic pathology, 2024: 19-43.
[15] 中国电子技术标准化研究院. 智慧家庭标准化白皮书[R], 2016.
[16] GB/T 25315—201. 家庭控制系统安全导则B/T 25315—201. 家庭控制系统安全导则[S], 2018.
[17] YD/T 6010—2024智能家居终端安全智能摄像头安全能力技术要求和测试方法[S], 2025.
[18] Jingwei Yi, Rui Ye, Qisi Chen.Robust Contrastive Language-Image Pre-training against Data Poisoning and Backdoor Attacks[A]// 37th Conference on Neural Information Processing Systems[C], 2023.
[19] GB/T 45288.1—2025人工智能大模型第1部分:通用要求[S], 2025.
[20] 中国信息通信研究院. 大模型基准测试体系研究报告(2024年)[R], 2024.
[21] Darren Edge, Ha Trinh, Newman Cheng, et al.From Local to Global: A Graph RAG Approach to Query-Focused Summarization[R]. Microsoft: CoRR abs/2404.16130, 2024.
[22] Joppe Bos, Leo Ducas, Eike Kiltz, et al.CRYSTALS - Kyber: A CCA-Secure Module-Lattice-Based KEM[A]// 2018 IEEE European Symposium on Security and Privacy[C], 2018:353-367.
[23] Daniel J. Bernstein, Andreas Hülsing, Stefan Kölbl, et al. The SPHINCS+ Signature Framework[A]// ACM SIGSAC Conference on Computer and Communications Security[C], 2019:2129-2146.