SaTC: CORE: Medium: Countering Surveillanceware Using Deception-Based Generative Models and Systems Mechanisms

监视软件(即跟踪软件、蠕虫软件、间谍软件等)是一种严重且日益普遍的网络安全威胁。在典型的情况下，恶意个人会在受害者的移动设备上安装软件，跟踪设备的位置，从而实现对其活动的远程监控。这不是假想的威胁：有报道称，亲密伴侣施虐者在受害者的智能手机上安装间谍软件，记者、持不同政见者和人权活动人士也同样成为专制政权的目标。杀毒软件等传统防御手段无法完全对抗这一威胁。虽然杀毒软件可能能够标记和删除监控软件，但由于人身暴力威胁等胁迫，一些受害者无法卸载监控软件。该项目旨在系统地研究监控软件，并为其开发基于人工智能(AI)的新防御措施。通过这样做，该项目有助于扩大网络安全研究，以包括易受伤害的个人和群体(例如亲密伴侣暴力的幸存者)的关切，他们的网络安全需求在历史上往往被忽视。为了进行这个项目，调查人员计划组建一个不同的团队，并与当地组织(如家庭暴力庇护所)和国际合作伙伴(如反偷窃软件联盟)合作。这项研究的重点是设计方法和工具来减轻监控软件的威胁，特别是开发一个基于欺骗的系统，使用机器学习技术和系统安全机制来产生虚假但可信的(“合成”)数据，并将其提供给监控软件的监控机构，而不是真实数据。研究自然分为三个阶段，首先对监视软件及其能力进行全面分析，以达到对抗性建模的目的。第二个推动力建立在这一分析的基础上，以开发技术来创建可用作诱饵的虚假但可信的数据。这需要使用机器学习技术，特别是深度生成模型。最后的推力包括设计系统机制，这些机制可以与前一次推力中开发的机械相结合，以确保防御的完整性。在这样做的过程中，该项目将推动对监视软件的正式对抗模型、合成可信数据和可否认数据嵌入的技术以及与机器学习技术相结合以阻止监视的系统级机制的理解。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Surveillanceware (i.e., stalkerware, creepware, spyware, etc.) is a serious and increasingly common cybersecurity threat. In a typical scenario, a malicious individual installs software on a victim's mobile device that tracks the device's location, enabling remote monitoring of its activity. This is not a hypothetical threat: there are reports of intimate partner abusers installing spyware on their victims' smartphones and of journalists, political dissidents, and human rights activists being similarly targeted by repressive regimes. Traditional defenses such as antivirus software are unable to fully counter this threat. While antivirus software may be able to flag and remove surveillanceware, some victims are unable to uninstall surveillanceware because of coercion such as threats of physical violence. This project seeks to systematically study surveillanceware and develop new artificial intelligence (AI)-based defenses for it. In doing so, the project helps broaden cybersecurity research to include the concerns of vulnerable individuals and groups (e.g., survivors of intimate partner violence) whose cybersecurity needs have often historically been neglected. To pursue the project, the investigators plan to assemble a diverse team and collaborate with local organizations (e.g., domestic abuse shelters) and international partners (e.g., the Coalition Against Stalkerware).The focus of this research effort is the design of methods and tools to mitigate the threat of surveillanceware, and in particular, developing a deception-based system that uses machine learning techniques and system security mechanisms to produce fake but plausible ("synthetic") data to be fed to the monitoring apparatus of surveillanceware instead of the real data. The research is naturally organized into three thrusts, starting with a comprehensive analysis of surveillanceware and its capabilities for the purpose of adversarial modeling. The second thrust builds on this analysis to develop techniques to create fake but plausible data that can be used as decoy. This requires the use of machine learning techniques, specifically deep generative models. The final thrust involves designing system mechanisms that can be combined with the machinery developed in the previous thrust to ensure the integrity of the defense. In so doing, the project will move forward an understanding of formal adversarial models for surveillanceware, techniques for synthesizing plausible data and deniable data embedding, and system-level mechanisms that integrate with machine learning techniques to thwart surveillance.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.