Membership inference attacks (MIAs) against in-context learning (ICL) serve as essential tools for privacy risk assessment and intellectual property safeguarding due to the use of small, private datasets for adaptation. However, most MIAs against language models require unrealistic, internal access or risk triggering built-in security mechanisms. In this paper, we propose Falcon (Flexible Attack on Language Context via ObfuscatioN), the first task-aware MIA framework against text-only model APIs. Falcon fully exploits the complexity of text obfuscation techniques and leverages the model’s discrepancies in reconstructing obfuscated texts from seen versus unseen data as a strong membership signal, successfully bypassing application constraints and LLM safeguarding mechanisms. Through extensive experiments on six widely used LLMs, including four open-source models (Llama-2, Llama-3, Qwen-2.5, Ministral) and two commercial models (GPT-3.5, and GPT-4o-mini), five datasets from various domains for tasks including question answering, text classification and summarization, Falcon generally achieves over 95% attack success rates, significantly outperforming existing methods. An in-depth analysis of the impact of model scale shows that Falcon exploits a capacity-induced vulnerability, indicating that models with higher capabilities are more susceptible to our attack. Additionally, we explore three defense methods, highlighting role validation as a potential mechanism for safeguarding LLM privacy. We have open-sourced Falcon’s modular, extensible codebase to support future research.

Emerging mobile apps such as UHD video and AR/VR access diverse high-throughput hardware devices, e.g., video codecs, cameras, and image processors. However, today's mobile emulators exhibit poor performance when emulating these devices. We pinpoint the major reason to be the discrepancy between the guest's and host's memory architectures for hardware devices, i.e., the mobile guest's centralized memory on a system-on-chip (SoC) versus the PC/server's separated memory modules on individual hardware. Such a discrepancy makes the shared virtual memory (SVM) architecture of mobile emulators highly inefficient.

Production-oriented virtual enterprise is a more important form in the virtual enterprise, quality information management plays an irreplaceable role in production-oriented virtual enterprise. According to life-cycle theory of the virtual enterprise and the condition of information technology, combing with quality assurance model and started with customer relations, quality information management system of production-oriented virtual enterprise was built. It is conducive to identify, acquire and deal with quality information, in order to provide quality information management guarantee for production-oriented virtual enterprise.