Anonymous communication relies on encrypted packet formats that resist traffic analysis and ensure unlinkability. Sphinx, the current standard for mixnets, provides strong anonymity but relies on classical public key cryptography, making it vulnerable to attacks from a quantum computer. In this paper, we present Outfox, a simplified variant of Sphinx tailored for mixnets with fixed length routes and designed for post-quantum security. Outfox eliminates a full key exchange and introduces a compact, per-hop header structure, reducing both computational and communication costs. We formally define Outfox and prove its security in the Universal Composability (UC) framework. Our evaluation shows that Outfox retains strong anonymity guarantees while offering improved efficiency and adaptability to quantum-resistant cryptographic primitives.
Millions of users leverage generative pretrained transformer (GPT)-based language models developed by leading model providers for a wide range of tasks. To support enhanced user interaction and customization, many platforms-such as OpenAI-now enable developers to create and publish tailored model instances, known as custom GPTs, via dedicated repositories or application stores. These custom GPTs empower users to browse and interact with specialized applications designed to meet specific needs. However, as custom GPTs see growing adoption, concerns regarding their security vulnerabilities have intensified. Existing research on these vulnerabilities remains largely theoretical, often lacking empirical, large-scale, and statistically rigorous assessments of associated risks. In this study, we analyze 14,904 custom GPTs to assess their susceptibility to seven exploitable threats, such as roleplay-based attacks, system prompt leakage, phishing content generation, and malicious code synthesis, across various categories and popularity tiers within the OpenAI marketplace. We introduce a multi-metric ranking system to examine the relationship between a custom GPT's popularity and its associated security risks. Our findings reveal that over 95% of custom GPTs lack adequate security protections. The most prevalent vulnerabilities include roleplay-based vulnerabilities (96.51%), system prompt leakage (92.20%), and phishing (91.22%). Furthermore, we demonstrate that OpenAI's foundational models exhibit inherent security weaknesses, which are often inherited or amplified in custom GPTs. These results highlight the urgent need for enhanced security measures and stricter content moderation to ensure the safe deployment of GPT-based applications.
We propose a new mechanism for digital identity schemes called 'blind-regroup', which, given a credential, allows a querying authority to identify all the existing credentials created by the same user. Unlike traceability, blind-regroup does not identify the ground identity of the user. Blind-regroup allows the user to continue creating credentials, and past blind-regroup queries do not compromise the anonymity of credentials created in the future. Blind-regroup is developed in a setting that satisfies authority transparency; that means that authorities can ultimately be held accountable for abuses of their power. We prove the correctness of blind-regroup (that is, it returns all and only all of the existing credentials matching the given credential); and we prove its security (namely, credentials not captured by blind-regroup remain unlinkable, and credentials are unforgeable). Our approach requires highly specialised zero-knowedge proofs; to demonstrate feasibility we provide SNARK implementations for these proofs and a performance analysis.
Vision-Language Models (VLMs) are increasingly deployed in applications that interpret and generate information from visual and textual inputs. While powerful, these models pose emerging privacy risks. In this paper, we introduce the concept of privacy chains: structured narratives that emerge when adversaries aggregate outputs from VLMs across multiple images, often exposing sensitive information even when the individual outputs are seemingly innocuous. Using LangChain, an open-source orchestration framework, we show how identity-linked data extracted via both benign and targeted prompts can be compiled into detailed timelines of private behavior, significantly amplifying privacy threats. To systematically assess this risk, we develop a privacy leakage pipeline within the Visual Question Answering (VQA) framework and evaluate six open-source VLMs across three tailored datasets: Celebrity, Car, and Tattoo. Our analysis reveals substantial and model-dependent privacy leakage, even from general-purpose queries. To mitigate this threat, we propose ChainShield, a white-box adversarial defense that applies targeted, imperceptible perturbations to images. ChainShield reduces privacy-relevant outputs by redirecting VLM responses toward benign alternatives, while preserving image realism. Our experiments show that ChainShield substantially lowers privacy leakage across models and datasets, effectively disrupting the formation of privacy chains.
Traffic surveillance enhances road safety and urban planning, but raises concerns that captured footage can be used to violate the privacy of drivers, passengers, or pedestrians. While current work focuses on protecting privacy from machine identifiers, the ability of human observers to memorize and recognize vehicles remains unstudied. This paper explores which vehicle features contribute to human memory retention through a pipeline designed to estimate vehicle memorability based on attention modeling and object-level features. We adopt AMNet, an attention-based memorability prediction model, to analyze how simple visual attributes-size, position, and color-affect vehicle memorability. We quantify these aspects and identify that memorability can be altered by simple changes in attributes such as vehicle's angle of view and distance from the camera. These findings offer insights for designing privacy-aware surveillance systems that address human perception threats.
