Search for: All records

Deepfake speech represents a real and growing threat to systems and society. Many detectors have been created to aid in defense against speech deepfakes. While these detectors implement myriad methodologies, many rely on low-level fragments of the speech generation process. We hypothesize that breath, a higher-level part of speech, is a key component of natural speech and thus improper generation in deepfake speech is a performant discriminator. To evaluate this, we create a breath detector and leverage this against a custom dataset of online news article audio to discriminate between real/deepfake speech. Additionally, we make this custom dataset publicly available to facilitate comparison for future work. Applying our simple breath detector as a deepfake speech discriminator on in-the-wild samples allows for accurate classification (perfect 1.0 AUPRC and 0.0 EER on test data) across 33.6 hours of audio. We compare our model with the state-of-the-art SSL-wav2vec and Codecfake models and show that these complex deep learning model completely either fail to classify the same in-the-wild samples (0.72 AUPRC and 0.89 EER), or substantially lack in the computational and temporal performance compared to our methodology (37 seconds to predict a one minute sample with Codecfake vs. 0.3 seconds with our model) 

Cochlear implants (CIs) allow deaf and hard-ofhearing individuals to use audio devices, such as phones or voice assistants. However, the advent of increasingly sophisticated synthetic audio (i.e., deepfakes) potentially threatens these users. Yet, this population’s susceptibility to such attacks is unclear. In this paper, we perform the first study of the impact of audio deepfakes on CI populations. We examine the use of CI-simulated audio within deepfake detectors. Based on these results, we conduct a user study with 35 CI users and 87 hearing persons (HPs) to determine differences in how CI users perceive deepfake audio. We show that CI users can, similarly to HPs, identify text-to-speech generated deepfakes. Yet, they perform substantially worse for voice conversion deepfake generation algorithms, achieving only 67% correct audio classification. We also evaluate how detection models trained on a CI-simulated audio compare to CI users and investigate if they can effectively act as proxies for CI users. This work begins an investigation into the intersection between adversarial audio and CI users to identify and mitigate threats against this marginalized group. 

IMSI-Catchers allow parties other than cellular network providers to covertly track mobile device users. While the research community has developed many tools to combat this problem, current solutions focus on correlated behavior and are therefore subject to substantial false classifications. In this paper, we present a standards-driven methodology that focuses on the messages an IMSI-Catcher must use to cause mobile devices to provide their permanent identifiers. That is, our approach focuses on causal attributes rather than correlated ones. We systematically analyze message flows that would lead to IMSI exposure (most of which have not been previously considered in the research community), and identify 53 messages an IMSI- Catcher can use for its attack. We then perform a measurement study on two continents to characterize the ratio in which connections use these messages in normal operations. We use these benchmarks to compare against open-source IMSI-Catcher implementations and then observe anomalous behavior at a large- scale event with significant media attention. Our analysis strongly implies the presence of an IMSI-Catcher at said public event (p << 0.005), thus representing the first publication to provide evidence of the statistical significance of its findings. 

In response to the growing sophistication of censor- ship methods deployed by governments worldwide, the existence of open-source censorship measurement platforms has increased. Analyzing censorship data is challenging due to the data’s large size, diversity, and variability, requiring a comprehensive under- standing of the data collection process and applying established data analysis techniques for thorough information extraction. In this work, we develop a framework that is applicable across all major censorship datasets to continually identify changes in cen- sorship data trends and reveal potentially unreported censorship. Our framework consists of control charts and the Mann-Kendall trend detection test, originating from statistical process control theory, and we implement it on Censored Planet, GFWatch, the Open Observatory of Network Interference (OONI), and Tor data from Russia, Myanmar, China, Iran, T ¨ urkiye, and Pakistan from January 2021 through March 2023. Our study confirms results from prior studies and also identifies new events that we validate through media reports. Our correlation analysis reveals minimal similarities between censorship datasets. However, because our framework is applicable across all major censorship datasets, it significantly reduces the manual effort required to employ multiple datasets, which we further demonstrate by applying it to four additional Internet outage-related datasets. Our work thus provides a tool for continuously monitoring censorship activity and acts as a basis for developing more systematic, holistic, and in-depth analysis techniques for censorship data. 

Audio deepfakes represent a rising threat to trust in our daily communications. In response to this, the research community has developed a wide array of detection techniques aimed at preventing such attacks from deceiving users. Unfortunately, the creation of these defenses has generally overlooked the most important element of the system - the user themselves. As such, it is not clear whether current mechanisms augment, hinder, or simply contradict human classification of deepfakes. In this paper, we perform the first large-scale user study on deepfake detection. We recruit over 1,200 users and present them with samples from the three most widely-cited deepfake datasets. We then quantitatively compare performance and qualitatively conduct thematic analysis to motivate and understand the reasoning behind user decisions and differences from machine classifications. Our results show that users correctly classify human audio at significantly higher rates than machine learning models, and rely on linguistic features and intuition when performing classification. However, users are also regularly misled by pre-conceptions about the capabilities of generated audio (e.g., that accents and background sounds are indicative of humans). Finally, machine learning models suffer from significantly higher false positive rates, and experience false negatives that humans correctly classify when issues of quality or robotic characteristics are reported. By analyzing user behavior across multiple deepfake datasets, our study demonstrates the need to more tightly compare user and machine learning performance, and to target the latter towards areas where humans are less likely to successfully identify threats. 

Industry is increasingly adopting private 5G networks to securely manage their wireless devices in retail, manufacturing, natural resources, and healthcare. As with most technology sectors, open- source software is well poised to form the foundation of deployments, whether it is deployed directly or as part of well-maintained proprietary offerings. This paper seeks to examine the use of cryptography and secure randomness in open-source cellular cores. We design a set of 13 CodeQL static program analysis rules for cores written in both C/C++ and Go and apply them to 7 open-source cellular cores implementing 4G and 5G functionality. We identify two significant security vulnerabilities, including predictable generation of TMSIs and improper verification of TLS certificates, with each vulnerability affecting multiple cores. In identifying these flaws, we hope to correct implementations to fix downstream deployments and derivative proprietary projects.