arXiv subjects

Sandra Siby

Publications and source records attributed to Sandra Siby.

13 recordsLinked to original sources

arXiv2026

From Monoliths to Swarms: A Study of Attack Surface Evolution in the Transition to Multi-Agent Web Systems

Large Language Model (LLM)-based web agents are increasingly evolving from single-agent systems (SAS) to multi-agent systems (MAS). While MAS can lead to improved task performance by decomposing complex tasks across specialized sub-agents, such role decomposition introduces new structural attack surfaces that are absent in SAS. This expanded attack surface remains poorly understood and inadequately categorized. To address this, we propose a taxonomy to categorize attack vectors specific to web-based MAS, accounting for vulnerabilities introduced or amplified by the involvement of multiple agents. We further present a test-bed WebMASLab to analyze web agent security against a fully external, web-only adversary. To isolate the effect of architecture, we keep the user task, tool surface, and browser substrate fixed, and compare single- and multi-agent setups. We evaluate three adversarial scenarios, across three conditions (baseline, prompt-hardened, and reasoning-enabled), including a novel MAS-specific Telephone Loop attack that exploits cross-agent delegation to create cyclical task loops. The attack is inert against SAS but compromises MAS when powered by three of the four frontier models evaluated (Claude Sonnet 4.5, GPT-5.2, GPT-5.4), averaging 80% across them at baseline. Only the fourth model, Claude Sonnet 4.6, resists the attack with a 92% detection rate. For the rest, the detection is 0% at baseline, reaching 33% with prompt-hardening for one model. We also show that obvious defenses do not generalize; prompt-hardening collapses one model's ASR from 100% to 8% while providing only modest reduction to the others. Our findings demonstrate that the transition from single- to multi-agent web systems changes the security landscape. Role specialization may not only lead to performance optimization but also introduce new architectural risks that require further study and defenses.

cs.CR↗

arXiv2026

Auditing the Impact of Cross-Site Web Tracking on YouTube Political and Misinformation Recommendations

YouTube has today become the primary news source for many users, which raises concerns about the role its recommendation algorithm can play in the spread of misinformation and political polarization. Prior work in this area has mainly analyzed how recommendations evolve based on users' watch history within the platform. Nevertheless, recommendations can also depend on off-platform browsing activity that Google collects via trackers on news websites, a factor that has not been considered so far. To fill this gap, we propose a sock-puppet-based experimental framework that automatically interacts with news media articles and then collects YouTube recommendations to measure how cross-site tracking affects the political and misinformation content users see. Moreover, by running our audits in both tracking-permissive and tracking-restrictive browser environments, we assess whether common privacy-focused browsers can protect users from tracking-driven political and misinformation bubbles on YouTube.

cs.CY↗

arXiv2026

Reckless Designs and Broken Promises: Privacy Implications of Targeted Interactive Advertisements on Social Media Platforms

Popular social media platforms TikTok, Facebook and Instagram allow third-parties to run targeted advertising campaigns on sensitive attributes in-platform. These ads are interactive by default, meaning users can comment or ``react'' (e.g., ``like'', ``love'') to them. We find that this platform-level design choice creates a privacy loophole such that advertisers can view the profiles of those who interact with their ads, thus identifying individuals that fulfill certain targeting criteria. This behavior is in contradiction to the promises made by the platforms to hide user data from advertisers. We conclude by suggesting design modifications that could provide users with transparency about the consequences of ad interaction to protect against unintentional disclosure.

cs.CR↗

arXiv2025

An Early Experience with Confidential Computing Architecture for On-Device Model Protection

Deploying machine learning (ML) models on user devices can improve privacy (by keeping data local) and reduce inference latency. Trusted Execution Environments (TEEs) are a practical solution for protecting proprietary models, yet existing TEE solutions have architectural constraints that hinder on-device model deployment. Arm Confidential Computing Architecture (CCA), a new Arm extension, addresses several of these limitations and shows promise as a secure platform for on-device ML. In this paper, we evaluate the performance-privacy trade-offs of deploying models within CCA, highlighting its potential to enable confidential and efficient ML applications. Our evaluations show that CCA can achieve an overhead of, at most, 22% in running models of different sizes and applications, including image classification, voice recognition, and chat assistants. This performance overhead comes with privacy benefits; for example, our framework can successfully protect the model against membership inference attack by an 8.3% reduction in the adversary's success rate. To support further research and early adoption, we make our code and methodology publicly available.

cs.CR↗

arXiv2024

P4: Towards private, personalized, and Peer-to-Peer learning

Personalized learning is a proposed approach to address the problem of data heterogeneity in collaborative machine learning. In a decentralized setting, the two main challenges of personalization are client clustering and data privacy. In this paper, we address these challenges by developing P4 (Personalized Private Peer-to-Peer) a method that ensures that each client receives a personalized model while maintaining differential privacy guarantee of each client's local dataset during and after the training. Our approach includes the design of a lightweight algorithm to identify similar clients and group them in a private, peer-to-peer (P2P) manner. Once grouped, we develop differentially-private knowledge distillation for clients to co-train with minimal impact on accuracy. We evaluate our proposed method on three benchmark datasets (FEMNIST or Federated EMNIST, CIFAR-10 and CIFAR-100) and two different neural network architectures (Linear and CNN-based networks) across a range of privacy parameters. The results demonstrate the potential of P4, as it outperforms the state-of-the-art of differential private P2P by up to 40 percent in terms of accuracy. We also show the practicality of P4 by implementing it on resource constrained devices, and validating that it has minimal overhead, e.g., about 7 seconds to run collaborative training between two clients.

cs.LG↗

arXiv2024

SINBAD: Saliency-informed detection of breakage caused by ad blocking

Privacy-enhancing blocking tools based on filter-list rules tend to break legitimate functionality. Filter-list maintainers could benefit from automated breakage detection tools that allow them to proactively fix problematic rules before deploying them to millions of users. We introduce SINBAD, an automated breakage detector that improves the accuracy over the state of the art by 20%, and is the first to detect dynamic breakage and breakage caused by style-oriented filter rules. The success of SINBAD is rooted in three innovations: (1) the use of user-reported breakage issues in forums that enable the creation of a high-quality dataset for training in which only breakage that users perceive as an issue is included; (2) the use of 'web saliency' to automatically identify user-relevant regions of a website on which to prioritize automated interactions aimed at triggering breakage; and (3) the analysis of webpages via subtrees which enables fine-grained identification of problematic filter rules.

cs.CR↗

arXiv2024

GuaranTEE: Towards Attestable and Private ML with CCA

Machine-learning (ML) models are increasingly being deployed on edge devices to provide a variety of services. However, their deployment is accompanied by challenges in model privacy and auditability. Model providers want to ensure that (i) their proprietary models are not exposed to third parties; and (ii) be able to get attestations that their genuine models are operating on edge devices in accordance with the service agreement with the user. Existing measures to address these challenges have been hindered by issues such as high overheads and limited capability (processing/secure memory) on edge devices. In this work, we propose GuaranTEE, a framework to provide attestable private machine learning on the edge. GuaranTEE uses Confidential Computing Architecture (CCA), Arm's latest architectural extension that allows for the creation and deployment of dynamic Trusted Execution Environments (TEEs) within which models can be executed. We evaluate CCA's feasibility to deploy ML models by developing, evaluating, and openly releasing a prototype. We also suggest improvements to CCA to facilitate its use in protecting the entire ML deployment pipeline on edge devices.

cs.CR↗

arXiv2024

PURL: Safe and Effective Sanitization of Link Decoration

While privacy-focused browsers have taken steps to block third-party cookies and mitigate browser fingerprinting, novel tracking techniques that can bypass existing countermeasures continue to emerge. Since trackers need to share information from the client-side to the server-side through link decoration regardless of the tracking technique they employ, a promising orthogonal approach is to detect and sanitize tracking information in decorated links. To this end, we present PURL (pronounced purel-l), a machine-learning approach that leverages a cross-layer graph representation of webpage execution to safely and effectively sanitize link decoration. Our evaluation shows that PURL significantly outperforms existing countermeasures in terms of accuracy and reducing website breakage while being robust to common evasion techniques. PURL's deployment on a sample of top-million websites shows that link decoration is abused for tracking on nearly three-quarters of the websites, often to share cookies, email addresses, and fingerprinting information.

cs.CR↗

arXiv2023

COOKIEGRAPH: Understanding and Detecting First-Party Tracking Cookies

As third-party cookie blocking is becoming the norm in browsers, advertisers and trackers have started to use first-party cookies for tracking. We conduct a differential measurement study on 10K websites with third-party cookies allowed and blocked. This study reveals that first-party cookies are used to store and exfiltrate identifiers to known trackers even when third-party cookies are blocked. As opposed to third-party cookie blocking, outright first-party cookie blocking is not practical because it would result in major functionality breakage. We propose CookieGraph, a machine learning-based approach that can accurately and robustly detect first-party tracking cookies. CookieGraph detects first-party tracking cookies with 90.20% accuracy, outperforming the state-of-the-art CookieBlock approach by 17.75%. We show that CookieGraph is fully robust against cookie name manipulation while CookieBlock's acuracy drops by 15.68%. While blocking all first-party cookies results in major breakage on 32% of the sites with SSO logins, and CookieBlock reduces it to 10%, we show that CookieGraph does not cause any major breakage on these sites. Our deployment of CookieGraph shows that first-party tracking cookies are used on 93.43% of the 10K websites. We also find that first-party tracking cookies are set by fingerprinting scripts. The most prevalent first-party tracking cookies are set by major advertising entities such as Google, Facebook, and TikTok.

cs.CR↗

arXiv2022

You get PADDING, everybody gets PADDING! You get privacy? Evaluating practical QUIC website fingerprinting protections for the masses

Website fingerprinting (WF) is a well-know threat to users' web privacy. New internet standards, such as QUIC, include padding to support defenses against WF. Previous work only analyzes the effectiveness of defenses when users are behind a VPN. Yet, this is not how most users browse the Internet. In this paper, we provide a comprehensive evaluation of QUIC-padding-based defenses against WF when users directly browse the web. We confirm previous claims that network-layer padding cannot provide good protection against powerful adversaries capable of observing all traffic traces. We further demonstrate that such padding is ineffective even against adversaries with constraints on traffic visibility and processing power. At the application layer, we show that defenses need to be deployed by both first and third parties, and that they can only thwart traffic analysis in limited situations. We identify challenges to deploy effective WF defenses and provide recommendations to address them.

cs.CR↗

arXiv2021

WebGraph: Capturing Advertising and Tracking Information Flows for Robust Blocking

Millions of web users directly depend on ad and tracker blocking tools to protect their privacy. However, existing ad and tracker blockers fall short because of their reliance on trivially susceptible advertising and tracking content. In this paper, we first demonstrate that the state-of-the-art machine learning based ad and tracker blockers, such as AdGraph, are susceptible to adversarial evasions deployed in real-world. Second, we introduce WebGraph, the first graph-based machine learning blocker that detects ads and trackers based on their action rather than their content. By building features around the actions that are fundamental to advertising and tracking - storing an identifier in the browser, or sharing an identifier with another tracker - WebGraph performs nearly as well as prior approaches, but is significantly more robust to adversarial evasions. In particular, we show that WebGraph achieves comparable accuracy to AdGraph, while significantly decreasing the success rate of an adversary from near-perfect under AdGraph to around 8% under WebGraph. Finally, we show that WebGraph remains robust to a more sophisticated adversary that uses evasion techniques beyond those currently deployed on the web.

cs.CR↗

arXiv2019

Encrypted DNS --> Privacy? A Traffic Analysis Perspective

Virtually every connection to an Internet service is preceded by a DNS lookup which is performed without any traffic-level protection, thus enabling manipulation, redirection, surveillance, and censorship. To address these issues, large organizations such as Google and Cloudflare are deploying recently standardized protocols that encrypt DNS traffic between end users and recursive resolvers such as DNS-over-TLS (DoT) and DNS-over-HTTPS (DoH). In this paper, we examine whether encrypting DNS traffic can protect users from traffic analysis-based monitoring and censoring. We propose a novel feature set to perform the attacks, as those used to attack HTTPS or Tor traffic are not suitable for DNS' characteristics. We show that traffic analysis enables the identification of domains with high accuracy in closed and open world settings, using 124 times less data than attacks on HTTPS flows. We find that factors such as location, resolver, platform, or client do mitigate the attacks performance but they are far from completely stopping them. Our results indicate that DNS-based censorship is still possible on encrypted DNS traffic. In fact, we demonstrate that the standardized padding schemes are not effective. Yet, Tor -- which does not effectively mitigate traffic analysis attacks on web traffic -- is a good defense against DoH traffic analysis.

cs.CR↗

arXiv2017

IoTScanner: Detecting and Classifying Privacy Threats in IoT Neighborhoods

In the context of the emerging Internet of Things (IoT), a proliferation of wireless connectivity can be expected. That ubiquitous wireless communication will be hard to centrally manage and control, and can be expected to be opaque to end users. As a result, owners and users of physical space are threatened to lose control over their digital environments. In this work, we propose the idea of an IoTScanner. The IoTScanner integrates a range of radios to allow local reconnaissance of existing wireless infrastructure and participating nodes. It enumerates such devices, identifies connection patterns, and provides valuable insights for technical support and home users alike. Using our IoTScanner, we attempt to classify actively streaming IP cameras from other non-camera devices using simple heuristics. We show that our classification approach achieves a high accuracy in an IoT setting consisting of a large number of IoT devices. While related work usually focuses on detecting either the infrastructure, or eavesdropping on traffic from a specific node, we focus on providing a general overview of operations in all observed networks. We do not assume prior knowledge of used SSIDs, preshared passwords, or similar.

cs.CR↗