Retrieval-Augmented Generation (RAG) grounds large language models in external corpora, but implicit trust in retrieved documents creates a critical attack surface: PoisonedRAG shows that a handful of crafted passages can dominate dense retrieval and steer generation toward attacker-chosen answers. We present the Tri-Layer Sieve, a middleware defense that sanitizes retrieved evidence through cross-embedding-space clustering with an independent judge model, structural filtering of trigger-payload artifacts, and LLM consistency verification. The design exploits a key weakness of retrieval-stage poisoning: a single document must satisfy one embedding geometry, one internal Trigger-Payload structure, and one generation objective - rarely all three simultaneously, a fragility that persists even against an adaptive attacker who paraphrases around it. On Natural Questions, HotpotQA, and MS-MARCO with Contriever retrieval (k=50), the Sieve reduces black-box Attack Success Rate from 67.0/87.0/64.0% to 3.0/14.0/4.0%, mitigates white-box HotFlip attacks from ~74% to 27.8% on NQ with Layer 3 enabled, and drives poisoned-document MRR to 0.000, while restoring clean accuracy from 13-33% under attack to 58-76%. Under an architecture-aware adversary who paraphrases triggers to evade the structural filter, enabling the consistency layer halves adaptive ASR (32.0% to 15.0% on NQ) while raising clean accuracy by 18 points, at an added latency of ~16-19 s/query under live retrieval.
Muhaimin Bin Munir, Akib Jawad Ononto, Nazia Shehnaz Joynab et al.· 0 citations
Privacy-aware Cloud-IoT architectures, such as DataBank, allow data owners to define their own privacy policies to control data sharing, however, these systems have not been designed specifically for smart home environments. In the context of Smart Home IoT simplicity is key, as user studies show that users seldom engage with lengthy policies, despite being concerned about data privacy. The main contributions of this paper are twofold: (i) the development of a data-sharing model for Smart Home IoT, and (ii) the design of a graph-based policy framework to govern interactions among data, devices, and services within this context. We chose a category-based model for policy definition since categorisation is a natural cognitive mechanism. To validate the proposed model, we implemented a prototype that illustrates how graph-based policies facilitate the visualisation and analysis of smart home policies. The prototype’s user interface enables individuals who are not security experts to create personalised privacy policies by specifying a limited set of preference settings. Participants in a user study were asked to interact with the prototype and subsequently complete a structured questionnaire to evaluate its usability and performance. Approximately 90% of participants found the tool helpful in configuring privacy policies and expressed satisfaction with its overall performance. The feature most appreciated by users was the policy analysis function, which aids in understanding the implications of selected privacy preferences. In conclusion, the proposed data sharing model and tool demonstrate potential in assisting users with the creation and interpretation of privacy policies in smart homes. While the focus of this paper is on Cloud and IoT-based Smart Homes, the concepts and technologies can be applied to various cyber physical systems such as Smart Transportation Systems and Smart Digital Healthcare Systems.
Weerasak Chiangsoi, Maribel Fernández, Jenjira Jaimunk et al.· IEEE International Conferenc...· 0 citations