NDSS 2025 – SKILLPoV: Towards Accessible And Effective Privacv Notice For Amazon Alexa Skills

Why It Matters

The solution bridges a critical accessibility gap, enabling users to make informed privacy decisions through the same conversational interface they use daily. It also sets a precedent for privacy‑by‑design in voice‑first ecosystems, influencing future regulations and developer practices.

Key Takeaways

• Voice privacy notices improve accessibility for visually impaired Alexa users
• SKILLPoV achieves 91.3% accuracy detecting data collection
• Generated notices use ChatGPT for concise, understandable content
• 92.3% of study participants liked the voice notice design
• Developers report seamless integration without altering skill functionality

Pulse Analysis

Privacy transparency has long been a stumbling block for voice assistants, where traditional web‑based policies are inaccessible to users who interact primarily through speech. The constrained screen‑less interface of devices like Amazon Echo forces users to seek privacy information on separate apps or browsers, a process that excludes visually impaired individuals and adds friction for all. By embedding concise, spoken notices directly into the skill activation flow, SKILLPoV redefines how privacy disclosures can be delivered, aligning the notice medium with the primary interaction channel.

Technically, SKILLPoV combines static code analysis with large‑language‑model generation to automate notice creation. The tool parses an Alexa skill’s source to identify data‑collection APIs, achieving 91.3% detection accuracy and 96.4% completeness, then prompts ChatGPT to craft short, user‑friendly scripts. These scripts are instrumented back into the skill without altering its core functionality, a claim validated by developer feedback during real‑world trials. This pipeline not only reduces manual effort for developers but also ensures that privacy language stays current with evolving data practices.

The broader impact extends beyond Alexa. As voice‑first platforms proliferate, regulators and industry groups are scrutinizing how privacy information is conveyed in non‑visual contexts. SKILLPoV offers a scalable, compliance‑friendly model that could be adapted for Google Assistant, Samsung Bixby, or emerging IoT voice agents. By proving that privacy notices can be both accessible and effective, the project encourages a shift toward privacy‑by‑design principles in conversational AI, fostering greater user trust and opening new avenues for inclusive technology design.