Navigating the Privacy Landscape of AI on the Devices: Challenges and Best Practices
summary
On-device generative AI introduces concrete risks: models ingest camera, microphone, motion sensors data used for recorder summarization, visual assistants and activity-recall snapshots, and can be accessed across phones and watches or by multiple users. Preinstalled models limit user control, updates and vulnerability fixes are hard because devices vary, and fine-tuning or sync can leak data. Mitigations include in-app permissions, model-access logging, extended retention/storage policies, adversarial testing, and engineering work on privacy-preserving fine-tuning and AI Core model management features.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:12Generative AI on devices
- 2:37Privacy risks of on-device AI
- 4:44Privacy risk assessment factors
- 6:52Fine-tuning and data handling
- 7:55Deployment and user control
- 9:18Model lifecycle challenges
- 10:42Privacy risk mitigation
- 12:25Technical solutions and future work
- 13:53Call to action
- 14:36Q&A
related talks
45mCombating Generative AI's Privacy Abuses
45mOverview of Risks, Vulnerabilities and Paths to Safer Use of Generative AI
01h 05mNavigating the Security and Privacy Landscape of Modern AI
27mSafeguarding Generative AI Applications: Navigating Security Challenges and Best Pract...
44mGenerative AI Security — A Practical Guide to Securing Your AI Application
30mPrivate Generative Models: Balancing Privacy and Utility in AI
13m"We are not Future-ready": Understanding AI Privacy Risks and Existing Mitigation...
31mSecurity Considerations for Services Using AI Models