Dangerous Skills Got Certified: Measuring the Trustworthiness of Skill Certification in Voice Personal Assistant Platforms
Long Cheng, Christin Wilson, Song Liao, Jeffrey Young, Daniel Dong, Hongxin Hu
Abstract
With the emergence of the voice personal assistant (VPA) ecosystem, third-party developers are allowed to build new voice-apps are called skills in the Amazon Alexa platform and actions in the Google Assistant platform, respectively. For the sake of brevity, we use the term skills to describe voice-apps including Amazon skills and Google actions, unless we need to distinguish them for different VPA platforms. and publish them to the skills store, which greatly extends the functionalities of VPAs. Before a new skill becomes publicly available, that skill must pass a certification process, which verifies that it meets the necessary content and privacy policies. The trustworthiness of skill certification is of significant importance to platform providers, developers, and end users. Yet, little is known about how difficult it is for a policy-violating skill to get certified and published in VPA platforms. In this work, we study the trustworthiness of the skill certification in Amazon Alexa and Google Assistant platforms to answer three key questions: 1) Whether the skill certification process is trustworthy in terms of catching policy violations in third-party skills. 2) Whether there exist policy-violating skills published in their skills stores. 3) What are VPA users' perspectives on the skill certification and their vulnerable usage behavior when interacting with VPA devices? Over a span of 15 months, we crafted and submitted for certification 234 Amazon Alexa skills and 381 Google Assistant actions that intentionally violate content and privacy policies specified by VPA platforms. Surprisingly, we successfully got 234 (100%) policy-violating Alexa skills certified and 148 (39%) policy-violating Google actions certified. Our analysis demonstrates that policy-violating skills exist in the current skills stores, and thus users (children, in particular) are at risk when using VPA services. We conducted a user study with 203 participants to understand users' misplaced trust on VPA platforms. Unfortunately, user expectations are not being met by the skill certification in leading VPA platforms.
Ask about this paper
Your agent reads all of it.
Lune indexed this paper to the last equation, along with the top-tier papers that cite it. Ask a question and the answer quotes them.
Your agent calls
Luneget_paper_fulltext
Free to start. No credit card required.
Terminal
Install the CLIlune papers fulltext b4e83616-409b-4cb3-9097-72191a47823eCited by top-tier papers14
- The Illusion of Empathy? Notes on Displays of Emotion in Human-Computer InteractionAndrea Cuadra, Maria Wang, Lynn Andrea Stein, Malte F. Jung et al.CHI 2024 · 75 citations
- P-Verifier: Understanding and Mitigating Security Risks in Cloud-based IoT Access PoliciesZe Jin, Luyi Xing, Yiwei Fang, Yan Jia et al.CCS 2022 · 19 citations
- VerHealth: Vetting Medical Voice Applications through Policy EnforcementFaysal Hossain Shezan, Hang Hu, Gang Wang, Yuan TianUbiComp 2021 · 18 citations
- Legal Obligation and Ethical Best Practice: Towards Meaningful Verbal Consent for Voice AssistantsWilliam Seymour, Mark Coté, Jose M. SuchCHI 2023 · 18 citations
- Demystifying the Vetting Process of Voice-controlled Skills on MarketsDawei Wang, Kai Chen, Wei WangUbiComp 2021 · 12 citations
Builds on11
- DolphinAttack: Inaudible Voice CommandsGuoming Zhang, Chen Yan, Xiaoyu Ji, Tianchen Zhang et al.CCS 2017 · 753 citations
- Hidden Voice CommandsNicholas Carlini, Pratyush Mishra, Tavish Vaidya, Yuankai Zhang et al.USENIX Security 2016 · 672 citations
- CommanderSong: A Systematic Approach for Practical Adversarial Voice RecognitionXuejing Yuan, Yuxuan Chen, Yue Zhao, Yunhui Long et al.USENIX Security 2018 · 389 citations
- Adversarial Attacks Against Automatic Speech Recognition Systems via Psychoacoustic HidingLea Schönherr, Katharina Kohls, Steffen Zeiler, Thorsten Holz et al.NDSS 2019 · 315 citations
- Who is Real Bob? Adversarial Attacks on Speaker Recognition SystemsGuangke Chen, Sen Chen, Lingling Fan, Xiaoning Du et al.S&P 2021 · 239 citations
Related papers
- SkillDetective: Automated Policy-Violation Detection of Voice Assistant Applications in the WildJeffrey Young, Song Liao, Long Cheng, Hongxin Hu et al.USENIX Security 2022
- SkillScanner: Detecting Policy-Violating Voice Applications Through Static Analysis at the Development PhaseSong Liao, Long Cheng, Haipeng Cai, Linke Guo et al.CCS 2023 · 7 citations
- Voice App Developer Experiences with Alexa and Google Assistant: Juggling Risks, Liability, and SecurityWilliam Seymour, Noura Abdi, Kopo M. Ramokapane, Jide S. Edu et al.USENIX Security 2024 · 9 citations
- Hey Alexa, is this Skill Safe?: Taking a Closer Look at the Alexa Skill EcosystemChristopher Lentzsch, Sheel Jayesh Shah, Benjamin Andow, Martin Degeling et al.NDSS 2021
- SkillExplorer: Understanding the Behavior of Skills in Large ScaleZhixiu Guo, Zijin Lin, Pan Li, Kai ChenUSENIX Security 2020
