This work focuses on the early ChatGPT plugin ecosystem, which is in the original plugin store before the transition to the GPT Store, and probes domain reachability, residual configuration files, and API endpoint responses to identify backend services that standard probes miss.
Abstract
As large language model (LLM) platforms evolve rapidly, third-party plugin systems extend model capabilities but also introduce challenges for infrastructure governance and security. When a platform replaces an older third-party plugin ecosystem, backend services from the old ecosystem may remain reachable even after the platform no longer exposes those integrations. In this work, we focus on the early ChatGPT plugin ecosystem, which is in the original plugin store before the transition to the GPT Store. We probe domain reachability, residual configuration files, and API endpoint responses. We also use historical API records to identify backend services that standard probes miss. Our results show that 72.9% of plugin domains remain reachable, while only 6.2% of plugins retain fully functional backend APIs. Among these reachable backends, many do not declare authentication and expose numerous API endpoints that perform stateful operations. Historical probing further reveals 123 hidden backend services that remain publicly reachable but no longer appear in current configurations. These findings show that platform evolution reduces backend visibility but does not eliminate network reachability. This gap reveals a governance blind spot during the evolution of LLM platform extension mechanisms.
This work presents the first privacy- and security-oriented measurement of M365 third-party applications, and finds that only 1,069 of them expose both descriptions and permission sets, with significant inconsistencies in transparency across official distribution channels.
Vincenzo Longo, Alberto Verna, Nikhil Jha et al.· 0 citations
Whether off-the-shelf Large Language Models (LLMs) can effectively reason about taint flows in Android apps is investigated, and preliminary findings suggest that LLM reasoning may effectively complement traditional static taint analysis.
Nicholas Miazzo, Marco Alecci, Jordan Samhi et al.· 0 citations
AppEval is presented, a benchmark and native-toolchain evaluation framework for mobile application repair across HarmonyOS/ArkTS, iOS/Swift, and Android/Kotlin, and shows that mobile repair performance depends strongly on the evaluated agent while demonstrating why runtime-aware acceptance is necessary for meaningful comparison.
LLM API resellers have become an important access layer to modern LLM services. However, multi-level resale creates an opaque supply chain: a user's request may traverse undisclosed upstream resellers, each of which can inspect or modify prompts and responses, inducing ecosystem-level confidentiality and integrity risks. Existing studies audit individual resellers, but provide little visibility into hidden dependencies across resellers. We present CacheTracer, the first API-only measurement of such hidden dependencies. Our key insight is to exploit prefix-cache reuse as a side channel to measure dependency via cache-reach relations. CacheTracer operationalizes this insight with two primitives: Flood populates fresh cache state through one endpoint, and Prove probes whether another can reuse it while excluding probe-created hits. We then conduct a real-world measurement study with CacheTracer on 39 reseller endpoints, sending 1.1 million API requests across 636 endpoint pairs. Our measurements reveal a deep, concentrated cache-reach structure: 37.1% of measured pairs exhibit shared cache reach, the containment order spans seven layers, and one cache reach is contained within at least 31 of other nodes. We further find that the recovered structure is model-specific. We also evaluate the validity of CacheTracer through both real-world consistency checks and controlled experiments. The results show its high reliability and accuracy. These findings reveal substantial hidden dependencies among seemingly independent API resellers. Such deep and concentrated dependencies can create a large potential blast radius, where a confidentiality or integrity failure along a common upstream path may affect users across multiple downstream resellers.
The Android Open Source Project (AOSP) powers the world's largest mobile ecosystem, yet memory-safety vulnerabilities remain prevalent in native components due to the inherent risks of languages like C and C++, which allow manual memory management and unsafe operations. To mitigate these risks, AOSP began adopting Rust, a language with compile-time memory-safety guarantees, in 2021. However, the trajectory, mechanisms, and effectiveness of this transition remain empirically uncharted. We present the first longitudinal and systematic empirical study of Rust in AOSP. By analyzing 16 quarterly snapshots from January 2021 to October 2024, we track adoption trends and map the distribution of Rust code across 67 components. We further align Android Security Bulletin vulnerabilities with these components to assess memory-safety outcomes before and after Rust adoption, while also identifying the remaining non-Rust hotspots that may warrant future migration. In addition, we characterize platform-level integration mechanisms and examine unsafe governance through a large corpus of SAFETY comments, from which we distill actionable best practices and anti-patterns. Taken together, these results clarify Rust's role in AOSP by shedding light on its adoption, integration, and governance in a safety-critical mobile platform. They also provide industry practitioners with an empirical foundation and practical guidance for similar memory-safe migrations in large-scale production systems.
This study investigates whether a frontier LLM can generate Dockerfiles and Docker Compose configurations for multi-service applications using repository contents without access to developer-authored deployment artifacts and analytically derives a minimal explicit deployment specification for information that cannot be reliably inferred from repository artifacts.
Oleg Grynets, Kyrylo Fursov, V. Lyashkevych et al.· 0 citations