Online scams continue to cause substantial financial and personal harm. As a result, detection systems based on Large Language Models (LLMs) have been integrated into security products ranging from email gateways and browser extensions to fraud-monitoring dashboards. As this adoption accelerates, a common belief has taken hold: that these models are broadly suitable for scam detection. In this work, we investigate whether LLMs, with their strong capabilities in understanding intent, context, and reasoning, can effectively detect scams across diverse scenarios without task-specific fine-tuning. We curate and release a unique benchmark dataset of real-world scams spanning multiple formats and topics. We evaluate nine LLMs of varying sizes and architectures, examining their performance under different prompting strategies and comparing them to a fine-tuned BERT-based classifier. Our results show that while larger LLMs generally outperform smaller ones, effective prompting substantially boosts the performance of smaller models. Moreover, LLMs are better at generalizing to unseen scams compared to fine-tuned models, suggesting that pre-trained knowledge contributes meaningfully to scam detection. We release our dataset and evaluation framework to facilitate future research in robust scam detection using language models.

Modern commercial antivirus systems increasingly rely on machine learning (ML) to keep up with the rampant inflation of new malware. However, it is well-known that machine learning models are vulnerable to adversarial examples (AEs). Previous works have shown that ML malware classifiers are fragile to the white-box adversarial attacks. However, ML models used in commercial antivirus (AV) products are usually not available to attackers and only return hard classification labels. Therefore, it is more practical to evaluate the robustness of ML models and real-world AVs in a pure black-box manner. We propose a black-box Reinforcement Learning (RL) based framework to generate AEs for PE malware classifiers and AV engines. It regards the adversarial attack problem as a multi-armed bandit problem, which finds an optimal balance between exploiting the successful patterns and exploring more varieties. Compared to other frameworks, our improvements lie in three points: 1) limiting the exploration space by modeling the generation process as a stateless process to avoid combination explosions, 2) reusing the successful payload in modeling; and 3) minimizing the changes on AE samples to correctly assign the rewards in RL learning (which also helps identify the root cause of evasions). As a result, our framework has much higher evasion rates than other off-the-shelf frameworks. Results show it has over 74%--97% evasion rate for two state-of-the-art ML detectors and over 32%--48% evasion rate for commercial AVs in a pure black-box setting. We also demonstrate that the transferability of adversarial attacks among ML-based classifiers is higher than that between ML-based classifiers and commercial AVs.

Recent years have seen a dramatic increase in applications of Artificial Intelligence (AI), Machine Learning (ML), and data mining to security and privacy problems. The analytic tools and intelligent behavior provided by these techniques make AI and ML increasingly important for autonomous real-time analysis and decision making in domains with a wealth of data or that require quick reactions to constantly changing situations. The use of learning methods in security-sensitive domains, in which adversaries may attempt to mislead or evade intelligent machines, creates new frontiers for security research. The recent widespread adoption of "deep learning'' techniques, whose security properties are difficult to reason about directly, has only added to the importance of this research. In addition, data mining and machine learning techniques create a wealth of privacy issues, due to the abundance and accessibility of data. The AISec workshop provides a venue for presenting and discussing new developments in the intersection of security and privacy with AI and machine learning.

Blocklists, consisting of known malicious IP addresses, can be used as a simple method to block malicious traffic. However, blocklists can potentially lead to unjust blocking of legitimate users due to IP address reuse, where more users could be blocked than intended. IP addresses can be reused either at the same time (Network Address Translation) or over time (dynamic addressing). We propose two new techniques to identify reused addresses. We built a crawler using the BitTorrent Distributed Hash Table to detect NATed addresses and use the RIPE Atlas measurement logs to detect dynamically allocated address spaces. We then analyze 151 publicly available IPv4 blocklists to show the implications of reused addresses and find that 53-60% of blocklists contain reused addresses having about 30.6K-45.1K listings of reused addresses. We also find that reused addresses can potentially affect as many as 78 legitimate users for as many as 44 days.

Recent years have seen a dramatic increase in applications of Artificial Intelligence (AI) and Machine Learning (ML) to security and privacy problems. The analytic tools and intelligent behavior provided by these techniques make AI and ML increasingly important for autonomous real-time analysis and decision making in domains with a wealth of data or that require quick reactions to constantly changing situations. The use of learning methods in security-sensitive domains, in which adversaries may attempt to mislead or evade intelligent machines, creates new frontiers for security research. The recent widespread adoption of deep-learning techniques, whose security properties are difficult to reason about directly, has only added to the importance of this research. In addition, data mining and machine learning techniques create a wealth of privacy issues, due to the abundance and accessibility of data. The 12th ACM Workshop on Artificial Intelligence and Security (AISec) is one of the historical, leading venues for presenting and discussing new developments in the intersection of security and privacy with AI and ML.

Greg Durrett, Jonathan K. Kummerfeld, Taylor Berg-Kirkpatrick, Rebecca Portnoff, Sadia Afroz, Damon McCoy, Kirill Levchenko, Vern Paxson. Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing. 2017.

Sites for online classified ads selling sex are widely used by human traffickers to support their pernicious business. The sheer quantity of ads makes manual exploration and analysis unscalable. In addition, discerning whether an ad is advertising a trafficked victim or an independent sex worker is a very difficult task. Very little concrete ground truth (i.e., ads definitively known to be posted by a trafficker) exists in this space. In this work, we develop tools and techniques that can be used separately and in conjunction to group sex ads by their true owner (and not the claimed author in the ad). Specifically, we develop a machine learning classifier that uses stylometry to distinguish between ads posted by the same vs. different authors with 90% TPR and 1% FPR. We also design a linking technique that takes advantage of leakages from the Bitcoin mempool, blockchain and sex ad site, to link a subset of sex ads to Bitcoin public wallets and transactions. Finally, we demonstrate via a 4-week proof of concept using Backpage as the sex ad site, how an analyst can use these automated approaches to potentially find human traffickers.

Underground forums are widely used by criminals to buy and sell a host of stolen items, datasets, resources, and criminal services. These forums contain important resources for understanding cybercrime. However, the number of forums, their size, and the domain expertise required to understand the markets makes manual exploration of these forums unscalable. In this work, we propose an automated, top-down approach for analyzing underground forums. Our approach uses natural language processing and machine learning to automatically generate high-level information about underground forums, first identifying posts related to transactions, and then extracting products and prices. We also demonstrate, via a pair of case studies, how an analyst can use these automated approaches to investigate other categories of products and transactions. We use eight distinct forums to assess our tools: Antichat, Blackhat World, Carders, Darkode, Hack Forums, Hell, L33tCrew and Nulled. Our automated approach is fast and accurate, achieving over 80% accuracy in detecting post category, product, and prices.

Effective evaluations of approaches to circumventing government Internet censorship require incorporating perspectives of how censors operate in practice. We undertake an extensive examination of real censors by surveying prior measurement studies and analyzing field reports and bug tickets from practitioners. We assess both deployed circumvention approaches and research proposals to consider the criteria employed in their evaluations and compare these to the observed behaviors of real censors, identifying areas where evaluations could more faithfully and effectively incorporate the practices of modern censors. These observations lead to an agenda realigning research with the predominant problems of today.

Recent studies on Website Fingerprinting (WF) claim to have found highly effective attacks on Tor. However, these studies make assumptions about user settings, adversary capabilities, and the nature of the Web that do not necessarily hold in practical scenarios. The following study critically evaluates these assumptions by conducting the attack where the assumptions do not hold. We show that certain variables, for example, user's browsing habits, differences in location and version of Tor Browser Bundle, that are usually omitted from the current WF model have a significant impact on the efficacy of the attack. We also empirically show how prior work succumbs to the base rate fallacy in the open-world scenario. We address this problem by augmenting our classification method with a verification step. We conclude that even though this approach reduces the number of false positives over 63\%, it does not completely solve the problem, which remains an open issue for WF attacks.

Stylometry is a method for identifying anonymous authors of anonymous texts by analyzing their writing style. While stylometric methods have produced impressive results in previous experiments, we wanted to explore their performance on a challenging dataset of particular interest to the security research community. Analysis of underground forums can provide key information about who controls a given bot network or sells a service, and the size and scope of the cybercrime underworld. Previous analyses have been accomplished primarily through analysis of limited structured metadata and painstaking manual analysis. However, the key challenge is to automate this process, since this labor intensive manual approach clearly does not scale. We consider two scenarios. The first involves text written by an unknown cybercriminal and a set of potential suspects. This is standard, supervised stylometry problem made more difficult by multilingual forums that mix l33t-speak conversations with data dumps. In the second scenario, you want to feed a forum into an analysis engine and have it output possible doppelgangers, or users with multiple accounts. While other researchers have explored this problem, we propose a method that produces good results on actual separate accounts, as opposed to data sets created by artificially splitting authors into multiple identities. For scenario 1, we achieve 77% to 84% accuracy on private messages. For scenario 2, we achieve 94% recall with 90% precision on blogs and 85.18% precision with 82.14% recall for underground forum users. We demonstrate the utility of our approach with a case study that includes applying our technique to the Carders forum and manual analysis to validate the results, enabling the discovery of previously undetected doppelganger accounts.

In digital forensics, questions often arise about the authors of documents: their identity, demographic background, and whether they can be linked to other documents. The field of stylometry uses linguistic features and machine learning techniques to answer these questions. While stylometry techniques can identify authors with high accuracy in non-adversarial scenarios, their accuracy is reduced to random guessing when faced with authors who intentionally obfuscate their writing style or attempt to imitate that of another author. While these results are good for privacy, they raise concerns about fraud. We argue that some linguistic features change when people hide their writing style and by identifying those features, stylistic deception can be recognized. The major contribution of this work is a method for detecting stylistic deception in written documents. We show that using a large feature set, it is possible to distinguish regular documents from deceptive documents with 96.6% accuracy (F-measure). We also present an analysis of linguistic features that can be modified to hide writing style.

The use of stylometry, authorship recognition through purely linguistic means, has contributed to literary, historical, and criminal investigation breakthroughs. Existing stylometry research assumes that authors have not attempted to disguise their linguistic writing style. We challenge this basic assumption of existing stylometry methodologies and present a new area of research: adversarial stylometry. Adversaries have a devastating effect on the robustness of existing classification methods. Our work presents a framework for creating adversarial passages including obfuscation , where a subject attempts to hide her identity, and imitation , where a subject attempts to frame another subject by imitating his writing style, and translation where original passages are obfuscated with machine translation services. This research demonstrates that manual circumvention methods work very well while automated translation methods are not effective. The obfuscation method reduces the techniques' effectiveness to the level of random guessing and the imitation attempts succeed up to 67% of the time depending on the stylometry technique used. These results are more significant given the fact that experimental subjects were unfamiliar with stylometry, were not professional writers, and spent little time on the attacks. This article also contributes to the field by using human subjects to empirically validate the claim of high accuracy for four current techniques (without adversaries). We have also compiled and released two corpora of adversarial stylometry texts to promote research in this field with a total of 57 unique authors. We argue that this field is important to a multidisciplinary approach to privacy, security, and anonymity.