As content generated by Large Language Model (LLM) has grown exponentially, the ability to accurately identify and fingerprint such text has become increasingly crucial. In this work, we introduce a novel black-box approach for fingerprinting LLMs, achieving an impressive 72% accuracy in identifying the correct family of models (Such as Llama, Mistral, Gemma, etc) among a lineup of LLMs. We present an evolutionary strategy that leverages the capabilities of one LLM to discover the most salient features for identifying other LLMs. Our method employs a unique "Hide and Seek" algorithm, where an Auditor LLM generates discriminative prompts, and a Detective LLM analyzes the responses to fingerprint the target models. This approach not only demonstrates the feasibility of LLM-driven model identification but also reveals insights into the semantic manifolds of different LLM families. By iteratively refining prompts through in-context learning, our system uncovers subtle distinctions between model outputs, providing a powerful tool for LLM analysis and verification. This research opens new avenues for understanding LLM behavior and has significant implications for model attribution, security, and the broader field of AI transparency.
Hide and Seek in Embedding Space: Geometry-based Steganography and Detection in Large Language Models
UCL Centre for Artificial Intelligence, University College London, UK（伦敦大学学院人工智能中心，大学学院伦敦）；School of Computing and Communications, Lancaster University, UK（兰卡斯特大学计算机与通讯学院）；Department of Informatics, University of Sussex, UK（苏塞克斯大学信息学院）；Centre for Psychedelic Research and Centre for Complexity Science, Imperial College London, UK（伦敦帝国学院迷幻研究与复杂科学中心）；Centre for Eudaimonia and Human Flourishing, University of Oxford, UK（牛津大学幸福与人类繁荣中心）
As the web moves away from stateful tracking, browser fingerprinting is becoming more prevalent. Unfortunately, existing approaches to detect browser fingerprinting do not take into account potential evasion tactics such as code obfuscation. To address this gap, we investigate the robustness of a state-of-the-art fingerprinting detection approach against various off-the-shelf obfuscation tools. Overall, we find that the combination of static and dynamic analysis is robust against different types of obfuscation. While some obfuscators are able to induce false negatives in static analysis, dynam…
With the growth of communication over computer networks, how to maintain the confidentiality and security of transmitted information have become some of the important issues. In order to transfer data securely to the destination without unwanted disclosure or damage, nature inspired hide and seek tricks such as, cryptography and Steganography are heavily in use. Just like the Chameleon and many other bio-species those change their body color and hide themselves in the background in order to protect them from external attacks, Cryptography and Steganography are techniques those are used to encr…
Online anonymity and privacy has been based on confusing the adversary by creating indistinguishable network elements. Tor is the largest and most widely deployed anonymity system, designed against realistic modern adversaries. Recently, researchers have managed to fingerprint Tor's circuits -- and hence the type of underlying traffic -- simply by capturing and analyzing traffic traces. In this work, we study the circuit fingerprinting problem, isolating it from website fingerprinting, and revisit previous findings in this model, showing that accurate attacks are possible even when the applica…
Attacks like Spectre abuse speculative execution, one of the key performance optimizations of modern CPUs. Recently, several testing tools have emerged to automatically detect speculative leaks in commercial (black-box) CPUs. However, the testing process is still slow, which has hindered in-depth testing campaigns, and so far prevented the discovery of new classes of leakage. In this paper, we identify the root causes of the performance limitations in existing approaches, and propose techniques to overcome these limitations. With these techniques, we improve the testing speed over the state-of…
Payment channels effectively move the transaction load off-chain thereby successfully addressing the inherent scalability problem most cryptocurrencies face. A major drawback of payment channels is the need to ``top up'' funds on-chain when a channel is depleted. Rebalancing was proposed to alleviate this issue, where parties with depleting channels move their funds along a cycle to replenish their channels off-chain. Protocols for rebalancing so far either introduce local solutions or compromise privacy. In this work, we present an opt-in rebalancing protocol that is both private and globally…
University of Oxford（牛津大学）；University of Cambridge（剑桥大学）；University of California, Los Angeles（加利福尼亚大学洛杉矶分校）；University of Waterloo（滑铁卢大学）；Amsterdam UMC（阿姆斯特丹大学医学中心）；Cambridge University Hospitals NHS Foundation Trust（剑桥大学医院NHS基金会信托）；Microsoft Research（微软研究院）；Alan Turing Institute（艾伦·图灵研究所）
The clinical time-series setting poses a unique combination of challenges to data modeling and sharing. Due to the high dimensionality of clinical time series, adequate de-identification to preserve privacy while retaining data utility is difficult to achieve using common de-identification techniques. An innovative approach to this problem is synthetic data generation. From a technical perspective, a good generative model for time-series data should preserve temporal dynamics, in the sense that new sequences respect the original relationships between high-dimensional variables across time. Fro…
Steganography is the science of hiding and communicating a secret message by embedding it in an innocent looking text such that the eavesdropper is unaware of its existence. Previously, attempts were made to establish steganography using quantum key distribution (QKD). Recently, it has been shown that such protocols are vulnerable to a certain steganalysis attack that can detect the presence of the hidden message and suppress the entire communication. In this work, we elaborate on the vulnerabilities of the original protocol which make it insecure against this detection attack. Further, we pro…
University of Kansas（堪萨斯大学）；University of Science and Technology of China（中国科学技术大学）；University of Missouri–Kansas City（密苏里大学堪萨斯城分校）
With the growing popularity of artificial intelligence and machine learning, a wide spectrum of attacks against deep learning models have been proposed in the literature. Both the evasion attacks and the poisoning attacks attempt to utilize adversarially altered samples to fool the victim model to misclassify the adversarial sample. While such attacks claim to be or are expected to be stealthy, i.e., imperceptible to human eyes, such claims are rarely evaluated. In this paper, we present the first large-scale study on the stealthiness of adversarial samples used in the attacks against deep lea…
Human mobility is highly predictable. Individuals tend to only visit a few locations with high frequency, and to move among them in a certain sequence reflecting their habits and daily routine. This predictability has to be taken into account in the design of location privacy preserving mechanisms (LPPMs) in order to effectively protect users when they continuously expose their position to location-based services (LBSs). In this paper, we describe a method for creating LPPMs that are customized for a user's mobility profile taking into account privacy and quality of service requirements. By co…
Logic locking protects an IC from threats such as piracy of design IP and unauthorized overproduction throughout the IC supply chain. Out of the several techniques proposed by the research community, provably-secure logic locking (PSLL) has acquired a foothold due to its algorithmic and provable-security guarantees. However, the security of these techniques is questioned by attackers that exploit the vulnerabilities arising from the hardware implementation. Such attacks (i) are predominantly specific to locking techniques and (ii) lack generality and scalability. This leads to a plethora of at…
We introduce a new type of cryptographic primitive that we call hiding fingerprinting. A (quantum) fingerprinting scheme translates a binary string of length $n$ to $d$ (qu)bits, typically $d\ll n$, such that given any string $y$ and a fingerprint of $x$, one can decide with high accuracy whether $x=y$. Classical fingerprinting schemes cannot hide information very well: a classical fingerprint of $x$ that guarantees error at most $ε$ necessarily reveals $Ω(\log(1/ epsilon))$ bits about $x$. We call a scheme hiding if it reveals $o(\log(1/ε))$ bits; accordingly, no classical scheme is hiding. F…
In this demonstration, we put ourselves in the place of a website manager who seeks to use browser fingerprinting for web authentication. The first step is to choose the attributes to implement among the hundreds that are available. To do so, we developed BrFAST, an attribute selection platform that includes FPSelect, an algorithm that rigorously selects the attributes according to a trade-off between security and usability. BrFAST is configured with a set of parameters for which we provide values for BrFAST to be usable as is. We notably include the resources to use two publicly available bro…
Due to the frequent unauthorized access by commercial drones to Critical Infrastructures (CIs) such as airports and oil refineries, the US-based Federal Avionics Administration (FAA) recently published a new specification, namely RemoteID. The aforementioned rule mandates that all Unmanned Aerial Vehicles (UAVs) have to broadcast information about their identity and location wirelessly to allow for immediate invasion attribution. However, the enforcement of such a rule poses severe concerns on UAV operators, especially in terms of location privacy and tracking threats, to name a few. Indeed, b…
University of Maryland（马里兰大学）；Max Planck Institute for Informatics（马克斯·普朗克信息学研究所）；CISPA Helmholtz Center for Information Security（CISPA亥姆霍兹信息安全中心）
Photorealistic image generation has reached a new level of quality due to the breakthroughs of generative adversarial networks (GANs). Yet, the dark side of such deepfakes, the malicious use of generated media, raises concerns about visual misinformation. While existing research work on deepfake detection demonstrates high accuracy, it is subject to advances in generation techniques and adversarial iterations on detection countermeasure techniques. Thus, we seek a proactive and sustainable solution on deepfake detection, that is agnostic to the evolution of generative models, by introducing ar…
