In recent years, high-performance video recording devices have become ubiquitous, posing an unprecedented challenge to preserving personal privacy. As a result, privacy-preserving video systems have been receiving increased attention. In this paper, we present a novel privacy-preserving video algorithm that uses semantic segmentation to identify regions of interest, which are then anonymized with an adaptive blurring algorithm. This algorithm addresses two of the most important shortcomings of existing solutions: it is multi-scale, meaning it can identify and uniformly anonymize objects of different scales in the same image, and it is class-generic, so it can be used to anonymize any class of objects of interest. We show experimentally that our algorithm achieves excellent anonymity while preserving meaning in the visual data processed.

Facial recognition tools are becoming exceptionally accurate in identifying people from images. However, this comes at the cost of privacy for users of online services with photo management (e.g. social media platforms). Particularly troubling is the ability to leverage unsupervised learning to recognize faces even when the user has not labeled their images. In this paper we propose Ulixes, a strategy to generate visually non-invasive facial noise masks that yield adversarial examples, preventing the formation of identifiable user clusters in the embedding space of facial encoders. This is applicable even when a user is unmasked and labeled images are available online. We demonstrate the effectiveness of Ulixes by showing that various classification and clustering methods cannot reliably label the adversarial examples we generate. We also study the effects of Ulixes in various black-box settings and compare it to the current state of the art in adversarial machine learning. Finally, we challenge the effectiveness of Ulixes against adversarially trained models and show that it is robust to countermeasures.

Blockchain technologies are becoming increasingly popular for financial and business solutions thanks to their many features and adaptability. However, their security is not assured. Of particular concern are majority attacks, which exploit the consensus algorithm used to resolve competing chains to modify past and present transactions on the blockchain. Despite numerous in-the-wild majority attacks, detecting in-progress attacks remains difficult, with most attacks remaining undetected for hours or days. to understand better these attacks and why they are so difficult to detect, in this paper we present a series of increasingly sophisticated attacks performed in an experimental environment and analyze the resulting network traffic. We then propose a possible detection mechanism based on the shared characteristics of these attacks. The implementation of such an algorithm is finally tested in a variety of experimental environments.

The exponential growth of data in the information age poses several threats to the privacy and safety of digital service users. Existing legislation, such as the GDPR in Europe and the CCPA in California, defines frameworks and guidelines to promote personal privacy but leaves freedom in the choice of means to achieve privacy. Data anonymization techniques remove information that can be used to identify individuals from the dataset, either through suppression, generalization, anatomization, permutation, or perturbation. Information suppression remains the most common, safe, and widely applicable anonymization method, though at a high data utility cost. In this paper, we argue that even information suppression may not be sufficient in some cases.We study the case of a dataset that describes the shopping habits of a grocery store’s customers. All identifiers and quasi-identifiers are removed from the dataset by suppression. However, by augmenting the data in a novel multi-step, iterative process, and building a neural network enriched with prior knowledge, we show that most sensitive information can be retrieved with an accuracy of 80%.

This paper introduces a novel approach to membership inference attacks (MIA) targeting stable diffusion computer vision models, specifically focusing on the highly sophisticated Stable Diffusion V2 by StabilityAI. MIAs aim to extract sensitive information about a model's training data, posing significant privacy concerns. Despite its advancements in image synthesis, our research reveals privacy vulnerabilities in the stable diffusion models' outputs. Exploiting this information, we devise a black-box MIA that only needs to query the victim model repeatedly. Our methodology involves observing the output of a stable diffusion model at different generative epochs and training a classification model to distinguish when a series of intermediates originated from a training sample or not. We propose numerous ways to measure the membership features and discuss what works best. The attack's efficacy is assessed using the ROC AUC method, demonstrating a 60\% success rate in inferring membership information. This paper contributes to the growing body of research on privacy and security in machine learning, highlighting the need for robust defenses against MIAs. Our findings prompt a reevaluation of the privacy implications of stable diffusion models, urging practitioners and developers to implement enhanced security measures to safeguard against such attacks.