Byte by Byte: Unmasking Browser Fingerprinting at the Function Level Using V8 Bytecode Transformers Conference: Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security; October 13–17, 2025; Taipei, Taiwan Proceedings of the 2025 ACM SIGSAC Conference on Computer and Com

Pouneh Nikkhah Bahrami, Dylan Cutler, Igor Bilogrevic · 2025

Read Byte by Byte: Unmasking Browser Fingerprinting at the Function Level Using V8 Bytecode Transformers Conference: Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security; October 13–17, 2025; Taipei, Taiwan Proceedings of the 2025 ACM SIGSAC Conference on Computer and Com by Pouneh Nikkhah Bahrami, Dylan Cutler, Igor Bilogrevic (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-09950.txt.

Introduction

Pouneh Nikkhah Bahrami and 2 co-authors catch browser fingerprinting at the level of single functions in a 2025 privacy paper. Fingerprinting scripts quietly identify a visitor across sites, and blocking whole scripts either misses disguised code or breaks pages people need. Their system, ByteDefender, reads the V8 engine's bytecode instead of surface code, so renaming tricks and address games cannot hide what a function truly does.

A transformer classifier trained offline on bytecode sequences spots fingerprinting behaviour function by function. Lightweight signatures drawn from that model let a browser match suspicious functions during compilation, before they run, adding only a few percent to page load time. Tested across the hundred thousand most visited sites, the design promises surgical blocking that leaves honest scripts alone.

Catalog record

CREATORPouneh Nikkhah Bahrami, Dylan Cutler, Igor Bilogrevic
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE93 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-09950

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Byte by Byte: Unmasking Browser Fingerprinting at the Function Level Using V8 Bytecode Transformers Conference: Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security; October 13–17, 2025; Taipei, Taiwan Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security (CCS ’25), October 13–17, 2025, Taipei, Taiwan DOI: 10.1145/3719027.3765158 ISBN: 979-8-4007-1525-9/2025/10 CCS: Security and privacy Privacy protections Pouneh Nikkhah Bahrami, Dylan Cutler, Igor Bilogrevic Abstract . Browser fingerprinting enables persistent cross-site user tracking via subtle techniques that often evade conventional defenses or cause website breakage when script-level blocking countermeasures are applied. Addressing these challenges requires detection methods offering both function-level precision to minimize breakage and inherent robustness against code obfuscation and URL manipulation. We introduce ByteDefender, the first system leveraging V8 engine bytecode to detect fingerprinting operations specifically at the JavaScript function level. A Transformer-based classifier, trained offline on bytecode sequences, accurately identifies functions exhibiting fingerprinting behavior. We develop and evaluate lightweight signatures derived from this model to enable low-overhead, on-device matching against function bytecode during compilation but prior to execution, which only adds a 4% (average) latency to the page load time. This mechanism facilitates targeted, real-time prevention of fingerprinting function execution, thereby preserving legitimate script functionality. Operating directly on bytecode ensures inherent resilience against common code obfuscation and URL-based evasion. Our evaluation on the top 100k websites demonstrates high detection accuracy at both function- and script-level, with substantial improvements over state-of-the-art AST-based methods, particularly in robustness against obfuscation. ByteDefender offers a practical framework for effective, precise, and robust fingerprinting mitigation. † † cc-license: by 1. Introduction Modern JavaScript APIs grant web applications powerful capabilities, from interactive graphics to real-time communication but also expose system details that can be exploited for browser fingerprinting . This technique involves gathering a combination of device and browser characteristics, often through subtle variations in API responses or rendering outputs (e.g., Canvas ( Mowery and Shacham 2012 ) , AudioContext ( Das et al. 2018 ) , fonts ( Cao et al. 2017 ; Vastel et al. 2018 ) ), to create stable and quasi-unique identifiers. While sites may use these techniques for safety-oriented purposes (e.g. anti-fraud, anti-abuse, bot detection) ( Bursztein et al. 2016 ; Durey et al. 2021 ; Laperdrix et al. 2020 ; Senol et al. 2024 ; Kalantari et al. 2024 ) and for enhancing security by identifying returning devices without traditional authentication, they can also be re-purposed for pervasive, cross-site user tracking ( Englehardt and Narayanan 2016 ; Liu et al. 2025 ; Pugliese et al. 2020 ; Vastel et al. 2018 ; Eckersley 2010 ; Iqbal et al. 2021 ) . This tracking capability, often operating statelessly and without clear user awareness or control ( Cameron 2018 ; Nick Doty 2025 ) , raises significant privacy concerns since it can bypass standard privacy-enhancing tools and user preferences ( Iqbal et al. 2021 ) . Mitigating the privacy risks associated with fingerprinting, without undermining its safety-oriented uses or breaking website functionality, remains an open challenge. Existing countermeasures often fall short, as fine-grained defenses (e.g. adding noise to API outputs ( Team 2020 ; Datta et al. 2018 ) , normalizing values ( Cao et al. 2017 ; tormitigations [n. d.] ; bravemitigations 2018 ) , or restricting API access ( webkitBlocking [n. d.] ; w3mitigations [n. d.] ) ) could inadvertently disrupt critical fingerprinting applications (e.g., security checks) or break essential website features ( Vastel et al. 2018 ) . Furthermore, scripts may circumvent such defenses ( Vastel et al. 2018 ) or embrace newly developed fingerprinting vectors ( Bahrami et al. 2021 ) . Coarse-grained methods, primarily URL-based filter lists ( Disconnect 2024 ; privacybadger [n. d.] ; Englehardt and Narayanan 2016 ) , struggle to differentiate between scripts using fingerprinting for cross-site tracking versus non-tracking purposes. They also suffer from delays in updates and are vulnerable against common evasion techniques like CNAME cloaking ( Dao et al. 2021 ; Dimova et al. 2021 ) , URL randomization ( Wang et al. 2016 ) , and code obfuscation ( Moog et al. 2021 ; Ngan et al. 2022 ; Skolka et al. 2019 ) . Critically, blocking entire scripts often leads to web breakage, especially problematic when scripts mix essential functions with fingerprinting capabilities ( Amjad et al. 2024 ) , and is ineffective against inline scripts ( Iqbal et al. 2021 ) . Machine learning-based methods aim to automate the identification of fingerprinting behaviors by leveraging both static and dynamic features of JavaScript code. Static analysis approaches using ASTs ( Rizzo 2018 ; van Zalingen and Haanen 2018 ; Iqbal et al. 2021 ) have demonstrated effectiveness but remain susceptible to obfuscation techniques ( Ngan et al. 2022 ; Skolka et al. 2019 ) . Recent work by Ghasemisharif and Polakis ( Ghasemisharif and Polakis 2023 ) utilized V8 bytecode to detect general tracking scripts at the script-level , showing resilience against obfuscation. Amjad et al. ( Amjad et al. 2024 ) porposed a function-level detection framework for tracking behavior using dynamic execution context graphs. However, these approaches encounter three key limitations in addressing the fingerprinting dilemma. First, script-level analysis ( Ghasemisharif and Polakis 2023 ; Iqbal et al. 2021 ) is too coarse to isolate tracking behaviors embedded within multi-purpose scripts, risking disruption of core functionality or failing to block trackers. Second, dynamic context analysis ( Amjad et al. 2024 ) , while finer-grained, can be complex and can introduce runtime overhead, making efficient pre-execution intervention difficult. Third, neither ( Amjad et al. 2024 ; Ghasemisharif and Polakis 2023 ) specifically isolates fingerprinting behaviors; instead, both conflate stateless (fingerprinting) and stateful (traditional) tracking techniques. Yet, these two forms of tracking differ fundamentally in their mechanisms and objectives, and a unified detection approach is likely to exhibit mixed effectiveness. We argue that addressing the privacy risks of fingerprinting requires a solution that is precise (targeting individual functions), robust (resilient to evasion), and proactive (acting before execution). We introduce ByteDefender, the first system using V8 engine bytecode for function-level classification specifically aimed at identifying fingerprinting functions prior to execution. By analyzing the bytecode of individual functions during JavaScript compilation, ByteDefender achieves fine granularity and inherits robustness against source-level obfuscation ( Ngan et al. 2022 ; Rozi et al. 2020 ) . Distinct from dynamic context methods ( Amjad et al. 2024 ; Iqbal et al. 2021 ) , ByteDefender focuses on the static bytecode patterns within functions, enabling lightweight, pre-execution detection. We employ a Transformer-based classifier ( Vaswani et al. 2017 ; Devlin 2018 ; Kenton and Toutanova 2019 ) , trained offline to recognize specific bytecode behaviors associated with fingerprinting techniques ( Englehardt and Narayanan 2016 ; Iqbal et al. 2021 ) . ByteDefender can identify fingerprinting functions for efficient on-device matching against known fingerprinting signatures, enabling it to preemptively block or alter the execution of fingerprinting functions and mitigate potential privacy leakage without disrupting entire scripts. This targeted approach minimizes web breakage compared to script-level methods ( Ghasemisharif and Polakis 2023 ; Iqbal et al. 2021 ) and allows core website functionality to proceed without disruption. ByteDefender provides a practical pathway to effectively mitigate the privacy risks associated with browser fingerprinting while

Read the complete text