Panel: Privacy Design Patterns for AI Systems: Threats and Protections
summary
Panelists lay out concrete defenses and gaps: build an inventory, tag data at collection, and run anomaly detection and red‑team/pen‑test campaigns to find leaks; deploy differential privacy or federated learning where feasible but note DP is hard to scale and federated learning is not foolproof; watch for brittle anonymization (many vendors rely on regexes) and surprising operational failures (some vendors reuse a single encryption key); follow threat models like Plot for AI, log model access, and design provable unlearning and opt‑out/erase workflows.
AI-written summary · about
chapters (13)
- 0:00Privacy panel introduction
- 3:22Privacy risks in AI/ML
- 8:01AI data governance challenges
- 10:53ML data governance components
- 13:40Detecting AI privacy incidents
- 20:44Harmful AI/ML privacy patterns
- 22:31Bias and fairness strategies
- 25:20Data anonymization effectiveness
- 27:33Privacy assurance in AI
- 31:35AI regulations update
- 33:41Evolving areas without clear solutions yet
- 37:35Call to action for engineers
- 43:11Closing remarks and Q&A
related talks
33mPanel: How Privacy Engineers Can Shape the Coming Wave of AI Governance
01h 20mUnlocking societal benefits of AI with privacy protective technologies.
01h 00mPANEL: Statistical Theory, Privacy and Data Analysis
01h 17mPrivacy and New Technologies
01h 15mDATA PROTECTION ENGINEERING: WHAT IS THE ROAD AHEAD?
01h 19mAligning Data Privacy and AI Governance
13m"We are not Future-ready": Understanding AI Privacy Risks and Existing Mitigation...
01h 11mBack to Basics How Data Protection Principles Apply to AI Systems