arXiv subjects

Hamed Haddadi

Publications and source records attributed to Hamed Haddadi.

At least 37 records · Page 2Linked to original sources

arXiv2026

SPILLage: Agentic Oversharing on the Web

LLM-powered agents are beginning to automate user's tasks across the open web, often with access to user resources such as emails and calendars. Unlike standard LLMs answering questions in a controlled ChatBot setting, web agents act "in the wild", interacting with third parties and leaving behind an action trace. Therefore, we ask the question: how do web agents handle user resources when accomplishing tasks on their behalf across live websites? In this paper, we formalize Natural Agentic Oversharing -- the unintentional disclosure of task-irrelevant user information through an agent trace of actions on the web. We introduce SPILLage, a framework that characterizes oversharing along two dimensions: channel (content vs. behavior) and directness (explicit vs. implicit). This taxonomy reveals a critical blind spot: while prior work focuses on text leakage, web agents also overshare behaviorally through clicks, scrolls, and navigation patterns that can be monitored. We benchmark 180 tasks on live e-commerce sites with ground-truth annotations separating task-relevant from task-irrelevant attributes. Across 1,080 runs spanning two agentic frameworks and three backbone LLMs, we demonstrate that oversharing is pervasive with behavioral oversharing dominates content oversharing by 5x. This effect persists -- and can even worsen -- under prompt-level mitigation. However, removing task-irrelevant information before execution improves task success by up to 17.9%, demonstrating that reducing oversharing improves task success. Our findings underscore that protecting privacy in web agents is a fundamental challenge, requiring a broader view of "output" that accounts for what agents do on the web, not just what they type. Our datasets and code are available at https://github.com/jrohsc/SPILLage.

cs.AI↗

arXiv2025

Privacy Practices of Browser Agents

This paper presents a systematic evaluation of the privacy behaviors and attributes of eight recent, popular browser agents. Browser agents are software that automate Web browsing using large language models and ancillary tooling. However, the automated capabilities that make browser agents powerful also make them high-risk points of failure. Both the kinds of tasks browser agents are designed to execute, along with the kinds of information browser agents are entrusted with to fulfill those tasks, mean that vulnerabilities in these tools can result in enormous privacy harm. This work presents a framework of five broad factors (totaling 15 distinct measurements) to measure the privacy risks in browser agents. Our framework assesses i. vulnerabilities in the browser agent's components, ii. how the browser agent protects against website behaviors, iii. whether the browser agent prevents cross-site tracking, iv. how the agent responds to privacy-affecting prompts, and v. whether the tool leaks personal information to sites. We apply our framework to eight browser agents and identify 30 vulnerabilities, ranging from disabled browser privacy features to "autocompleting" sensitive personal information in form fields. We have responsibly disclosed our findings, and plan to release our dataset and other artifacts.

cs.CR↗

arXiv2025

Robust Hallucination Detection in LLMs via Adaptive Token Selection

Hallucinations in large language models (LLMs) pose significant safety concerns that impede their broader deployment. Recent research in hallucination detection has demonstrated that LLMs' internal representations contain truthfulness hints, which can be harnessed for detector training. However, the performance of these detectors is heavily dependent on the internal representations of predetermined tokens, fluctuating considerably when working on free-form generations with varying lengths and sparse distributions of hallucinated entities. To address this, we propose HaMI, a novel approach that enables robust detection of hallucinations through adaptive selection and learning of critical tokens that are most indicative of hallucinations. We achieve this robustness by an innovative formulation of the Hallucination detection task as Multiple Instance (HaMI) learning over token-level representations within a sequence, thereby facilitating a joint optimisation of token selection and hallucination detection on generation sequences of diverse forms. Comprehensive experimental results on four hallucination benchmarks show that HaMI significantly outperforms existing state-of-the-art approaches.

cs.LG↗

arXiv2025

Benchmarking Ultra-Low-Power $μ$NPUs

Efficient on-device neural network (NN) inference offers predictable latency, improved privacy and reliability, and lower operating costs for vendors than cloud-based inference. This has sparked recent development of microcontroller-scale NN accelerators, also known as neural processing units ($μ$NPUs), designed specifically for ultra-low-power applications. We present the first comparative evaluation of a number of commercially-available $μ$NPUs, including the first independent benchmarks for multiple platforms. To ensure fairness, we develop and open-source a model compilation pipeline supporting consistent benchmarking of quantized models across diverse microcontroller hardware. Our resulting analysis uncovers both expected performance trends as well as surprising disparities between hardware specifications and actual performance, including certain $μ$NPUs exhibiting unexpected scaling behaviors with model complexity. This work provides a foundation for ongoing evaluation of $μ$NPU platforms, alongside offering practical insights for both hardware and software developers in this rapidly evolving space.

cs.LG↗

arXiv2025

Secure and Confidential Certificates of Online Fairness

The black-box service model enables ML service providers to serve clients while keeping their intellectual property and client data confidential. Confidentiality is critical for delivering ML services legally and responsibly, but makes it difficult for outside parties to verify important model properties such as fairness. Existing methods that assess model fairness confidentially lack either (i) reliability because they certify fairness with respect to a static set of data, and therefore fail to guarantee fairness in the presence of distribution shift or service provider malfeasance; and/or (ii) scalability due to the computational overhead of confidentiality-preserving cryptographic primitives. We address these problems by introducing online fairness certificates, which verify that a model is fair with respect to data received by the service provider online during deployment. We then present OATH, a deployably efficient and scalable zero-knowledge proof protocol for confidential online group fairness certification. OATH exploits statistical properties of group fairness via a cut-and-choose style protocol, enabling scalability improvements over baselines.

cs.CY↗

arXiv2025

AegisMCP: Online Graph Intrusion Detection for Tool-Augmented LLMs on Edge Devices

In this work, we study security of Model Context Protocol (MCP) agent toolchains and their applications in smart homes. We introduce AegisMCP, a protocol-level intrusion detector. Our contributions are: (i) a minimal attack suite spanning instruction-driven escalation, chain-of-tool exfiltration, malicious MCP server registration, and persistence; (ii) NEBULA-Schema (Network-Edge Behavioral Learning for Untrusted LLM Agents), a reusable protocol-level instrumentation that represents MCP activity as a streaming heterogeneous temporal graph over agents, MCP servers, tools, devices, remotes, and sessions; and (iii) a CPU-only streaming detector that fuses novelty, session-DAG structure, and attribute cues for near-real-time edge inference, with optional fusion of local prompt-guardrail signals. On an emulated smart-home testbed spanning multiple MCP stacks and a physical bench, AegisMCP achieves sub-second per-window model inference and end-to-end alerting. The latency of AegisMCP is consistently sub-second on Intel N150-class edge hardware, while outperforming traffic-only and sequence baselines; ablations confirm the importance of DAG and install/permission signals. We release code, schemas, and generators for reproducible evaluation.

cs.CR↗

arXiv2025

Client Clustering Meets Knowledge Sharing: Enhancing Privacy and Robustness in Personalized Peer-to-Peer Learning

The growing adoption of Artificial Intelligence (AI) in Internet of Things (IoT) ecosystems has intensified the need for personalized learning methods that can operate efficiently and privately across heterogeneous, resource-constrained devices. However, enabling effective personalized learning in decentralized settings introduces several challenges, including efficient knowledge transfer between clients, protection of data privacy, and resilience against poisoning attacks. In this paper, we address these challenges by developing P4 (Personalized, Private, Peer-to-Peer) -- a method designed to deliver personalized models for resource-constrained IoT devices while ensuring differential privacy and robustness against poisoning attacks. Our solution employs a lightweight, fully decentralized algorithm to privately detect client similarity and form collaborative groups. Within each group, clients leverage differentially private knowledge distillation to co-train their models, maintaining high accuracy while ensuring robustness to the presence of malicious clients. We evaluate P4 on popular benchmark datasets using both linear and CNN-based architectures across various heterogeneity settings and attack scenarios. Experimental results show that P4 achieves 5% to 30% higher accuracy than leading differentially private peer-to-peer approaches and maintains robustness with up to 30% malicious clients. Additionally, we demonstrate its practicality by deploying it on resource-constrained devices, where collaborative training between two clients adds only ~7 seconds of overhead.

cs.LG↗

arXiv2025

An Architecture for Spatial Networking

Physical spaces are increasingly dense with networked devices, promising seamless coordination and ambient intelligence. Yet today, cloud-first architectures force all communication through wide-area networks regardless of physical proximity. We lack an abstraction for spatial networking: using physical spaces to create boundaries for private, robust, and low-latency communication. We introduce $\textit{Bifröst}$, a programming model that realizes spatial networking using bigraphs to express both containment and connectivity, enabling policies to be scoped by physical boundaries, devices to be named by location, the instantiation of spatial services, and the composition of spaces while maintaining local autonomy. Bifröst enables a new class of spatially-aware applications, where co-located devices communicate directly, physical barriers require explicit gateways, and local control bridges to global coordination.

cs.NI↗

arXiv2025

GuardNet: Graph-Attention Filtering for Jailbreak Defense in Large Language Models

Large Language Models (LLMs) are increasingly susceptible to jailbreak attacks, which are adversarial prompts that bypass alignment constraints and induce unauthorized or harmful behaviors. These vulnerabilities undermine the safety, reliability, and trustworthiness of LLM outputs, posing critical risks in domains such as healthcare, finance, and legal compliance. In this paper, we propose GuardNet, a hierarchical filtering framework that detects and filters jailbreak prompts prior to inference. GuardNet constructs structured graphs that combine sequential links, syntactic dependencies, and attention-derived token relations to capture both linguistic structure and contextual patterns indicative of jailbreak behavior. It then applies graph neural networks at two levels: (i) a prompt-level filter that detects global adversarial prompts, and (ii) a token-level filter that pinpoints fine-grained adversarial spans. Extensive experiments across three datasets and multiple attack settings show that GuardNet substantially outperforms prior defenses. It raises prompt-level F$_1$ scores from 66.4\% to 99.8\% on LLM-Fuzzer, and from 67-79\% to over 94\% on PLeak datasets. At the token level, GuardNet improves F$_1$ from 48-75\% to 74-91\%, with IoU gains up to +28\%. Despite its structural complexity, GuardNet maintains acceptable latency and generalizes well in cross-domain evaluations, making it a practical and robust defense against jailbreak threats in real-world LLM deployments.

cs.LG↗

arXiv2025

Context-Aware Membership Inference Attacks against Pre-trained Large Language Models

Membership Inference Attacks (MIAs) on pre-trained Large Language Models (LLMs) aim at determining if a data point was part of the model's training set. Prior MIAs that are built for classification models fail at LLMs, due to ignoring the generative nature of LLMs across token sequences. In this paper, we present a novel attack on pre-trained LLMs that adapts MIA statistical tests to the perplexity dynamics of subsequences within a data point. Our method significantly outperforms prior approaches, revealing context-dependent memorization patterns in pre-trained LLMs.

cs.CL↗

arXiv2025

Membership and Memorization in LLM Knowledge Distillation

Recent advances in Knowledge Distillation (KD) aim to mitigate the high computational demands of Large Language Models (LLMs) by transferring knowledge from a large ''teacher'' to a smaller ''student'' model. However, students may inherit the teacher's privacy when the teacher is trained on private data. In this work, we systematically characterize and investigate membership and memorization privacy risks inherent in six LLM KD techniques. Using instruction-tuning settings that span seven NLP tasks, together with three teacher model families (GPT-2, LLAMA-2, and OPT), and various size student models, we demonstrate that all existing LLM KD approaches carry membership and memorization privacy risks from the teacher to its students. However, the extent of privacy risks varies across different KD techniques. We systematically analyse how key LLM KD components (KD objective functions, student training data and NLP tasks) impact such privacy risks. We also demonstrate a significant disagreement between memorization and membership privacy risks of LLM KD techniques. Finally, we characterize per-block privacy risk and demonstrate that the privacy risk varies across different blocks by a large margin.

cs.LG↗

arXiv2025

REAL-IoT: Characterizing GNN Intrusion Detection Robustness under Practical Adversarial Attack

Graph Neural Network (GNN)-based network intrusion detection systems (NIDS) are often evaluated on single datasets, limiting their ability to generalize under distribution drift. Furthermore, their adversarial robustness is typically assessed using synthetic perturbations that lack realism. This measurement gap leads to an overestimation of GNN-based NIDS resilience. To address the limitations, we propose \textbf{REAL-IoT}, a comprehensive framework for robustness evaluation of GNN-based NIDS in IoT environments. Our framework presents a methodology that creates a unified dataset from canonical datasets to assess generalization under drift. In addition, it features a novel intrusion dataset collected from a physical IoT testbed, which captures network traffic and attack scenarios under real-world settings. Furthermore, using REAL-IoT, we explore the usage of Large Language Models (LLMs) to analyze network data and mitigate the impact of adversarial examples by filtering suspicious flows. Our evaluations using REAL-IoT reveal performance drops in GNN models compared to results from standard benchmarks, quantifying their susceptibility to drift and realistic attacks. We also demonstrate the potential of LLM-based filtering to enhance robustness. These findings emphasize the necessity of realistic threat modeling and rigorous measurement practices for developing resilient IoT intrusion detection systems.

cs.CR↗

arXiv2025

Measuring the Accuracy and Effectiveness of PII Removal Services

This paper presents the first large-scale empirical study of commercial personally identifiable information (PII) removal systems -- commercial services that claim to improve privacy by automating the removal of PII from data broker's databases. Popular examples of such services include DeleteMe, Mozilla Monitor, Incogni, among many others. The claims these services make may be very appealing to privacy-conscious Web users, but how effective these services actually are at improving privacy has not been investigated. This work aims to improve our understanding of commercial PII removal services in multiple ways. First, we conduct a user study where participants purchase subscriptions from four popular PII removal services, and report (i) what PII the service find, (ii) from which data brokers, (iii) whether the service is able to have the information removed, and (iv) whether the identified information actually is PII describing the participant. And second, by comparing the claims and promises the services makes (e.g. which and how many data brokers each service claims to cover). We find that these services have significant accuracy and coverage issues that limit the usefulness of these services as a privacy-enhancing technology. For example, we find that the measured services are unable to remove the majority of the identified PII records from data broker's (48.2% of the successfully removed found records) and that most records identified by these services are not PII about the user (study participants found that only 41.1% of records identified by these services were PII about themselves).

cs.CR↗

arXiv2025

Nebula: Efficient, Private and Accurate Histogram Estimation

We present \textit{Nebula}, a system for differentially private histogram estimation on data distributed among clients. \textit{Nebula} allows clients to independently decide whether to participate in the system, and locally encode their data so that an untrusted server only learns data values whose multiplicity exceeds a predefined aggregation threshold, with $(\varepsilon,δ)$ differential privacy guarantees. Compared to existing systems, \textit{Nebula} uniquely achieves: \textit{i)} a strict upper bound on client privacy leakage; \textit{ii)} significantly higher utility than standard local differential privacy systems; and \textit{iii)} no requirement for trusted third-parties, multi-party computation, or trusted hardware. We provide a formal evaluation of \textit{Nebula}'s privacy, utility and efficiency guarantees, along with an empirical assessment on three real-world datasets. On the United States Census dataset, clients can submit their data in just 0.0036 seconds and 0.0016 MB (\textbf{efficient}), under strong $(\varepsilon=1,δ=10^{-8})$ differential privacy guarantees (\textbf{private}), enabling \textit{Nebula}'s untrusted aggregation server to estimate histograms with over 88\% better utility than existing local differential privacy deployments (\textbf{accurate}). Additionally, we describe a variant that allows clients to submit multi-dimensional data, with similar privacy, utility, and performance. Finally, we provide an implementation of \textit{Nebula}.

cs.CR↗

arXiv2025

Local Frames: Exploiting Inherited Origins to Bypass Content Blockers

We present a study of how local frames (i.e., iframes loading content like "about:blank") are mishandled by a wide range of popular Web security and privacy tools. As a result, users of these tools remain vulnerable to the very attack techniques against which they seek to protect themselves, including browser fingerprinting, cookie-based tracking, and data exfiltration. The tools we study are vulnerable in different ways, but all share a root cause: legacy Web functionality interacts with browser privacy boundaries in unexpected ways, leading to systemic vulnerabilities in tools developed, maintained, and recommended by privacy experts and activists. We consider four core capabilities supported by most privacy tools and develop tests to determine whether each can be evaded through the use of local frames. We apply our tests to six popular Web privacy and security tools -- identifying at least one vulnerability in each for a total of 19 -- and extract common patterns regarding their mishandling of local frames. Our measurement of popular websites finds that 56% employ local frames and that 73.7% of the requests made by these local frames should be blocked by popular filter lists but instead trigger the vulnerabilities we identify. From another perspective, 14.3% of all sites that we crawl make requests that should be blocked inside of local frames. We disclosed these vulnerabilities to the tool authors and discuss both our experiences working with them to patch their products and the implications of our findings for other privacy and security research.

cs.CR↗

arXiv2025

Deep Unlearn: Benchmarking Machine Unlearning for Image Classification

Machine unlearning (MU) aims to remove the influence of particular data points from the learnable parameters of a trained machine learning model. This is a crucial capability in light of data privacy requirements, trustworthiness, and safety in deployed models. MU is particularly challenging for deep neural networks (DNNs), such as convolutional nets or vision transformers, as such DNNs tend to memorize a notable portion of their training dataset. Nevertheless, the community lacks a rigorous and multifaceted study that looks into the success of MU methods for DNNs. In this paper, we investigate 18 state-of-the-art MU methods across various benchmark datasets and models, with each evaluation conducted over 10 different initializations, a comprehensive evaluation involving MU over 100K models. We show that, with the proper hyperparameters, Masked Small Gradients (MSG) and Convolution Transpose (CT), consistently perform better in terms of model accuracy and run-time efficiency across different models, datasets, and initializations, assessed by population-based membership inference attacks (MIA) and per-sample unlearning likelihood ratio attacks (U-LiRA). Furthermore, our benchmark highlights the fact that comparing a MU method only with commonly used baselines, such as Gradient Ascent (GA) or Successive Random Relabeling (SRL), is inadequate, and we need better baselines like Negative Gradient Plus (NG+) with proper hyperparameter selection.

cs.LG↗

arXiv2025

Poster: Enhancing GNN Robustness for Network Intrusion Detection via Agent-based Analysis

Graph Neural Networks (GNNs) show great promise for Network Intrusion Detection Systems (NIDS), particularly in IoT environments, but suffer performance degradation due to distribution drift and lack robustness against realistic adversarial attacks. Current robustness evaluations often rely on unrealistic synthetic perturbations and lack demonstrations on systematic analysis of different kinds of adversarial attack, which encompass both black-box and white-box scenarios. This work proposes a novel approach to enhance GNN robustness and generalization by employing Large Language Models (LLMs) in an agentic pipeline as simulated cybersecurity expert agents. These agents scrutinize graph structures derived from network flow data, identifying and potentially mitigating suspicious or adversarially perturbed elements before GNN processing. Our experiments, using a framework designed for realistic evaluation and testing with a variety of adversarial attacks including a dataset collected from physical testbed experiments, demonstrate that integrating LLM analysis can significantly improve the resilience of GNN-based NIDS against challenges, showcasing the potential of LLM agent as a complementary layer in intrusion detection architectures.

cs.CR↗

arXiv2025

Verifiable Unlearning on Edge

Machine learning providers commonly distribute global models to edge devices, which subsequently personalize these models using local data. However, issues such as copyright infringements, biases, or regulatory requirements may require the verifiable removal of certain data samples across all edge devices. Ensuring that edge devices correctly execute such unlearning operations is critical to maintaining integrity. In this work, we introduce a verification framework leveraging zero-knowledge proofs, specifically zk-SNARKs, to confirm data unlearning on personalized edge-device models without compromising privacy. We have developed algorithms explicitly designed to facilitate unlearning operations that are compatible with efficient zk-SNARK proof generation, ensuring minimal computational and memory overhead suitable for constrained edge environments. Furthermore, our approach carefully preserves personalized enhancements on edge devices, maintaining model performance post-unlearning. Our results affirm the practicality and effectiveness of this verification framework, demonstrating verifiable unlearning with minimal degradation in personalization-induced performance improvements. Our methodology ensures verifiable, privacy-preserving, and effective machine unlearning across edge devices.

cs.LG↗