Developing AI in a privacy conscious world
summary
Claims concrete safeguards: no-training-on-customer-content and a three‑pillar approach — federated models (OpenAI, Infrric and others) plus responsible AI and empowerment. Demonstrated attack: Nightshade data poisoning flips labels (cat→cow), with 50–100 poison samples causing wrong outputs like dog→cat or handbag→toaster. Operational controls include ISO 27001‑style checks, pentests, red teams and a Zoom Video SDK prototype; timelines note EU AI Act operational around Q2 2026.
AI-written summary · about
chapters (11)
- 0:00Privacy conscious AI intro
- 1:19Generative AI privacy concerns
- 3:30AI for empowerment and art
- 4:34Trust and ethics in AI
- 6:02Zoom’s federated AI approach
- 7:48AI challenges and security
- 9:57Global AI legislation overview
- 12:07Responsible AI frameworks
- 13:52AI bias and data leaks
- 16:39Trust via transparency
- 17:21Privacy conscious AI principles
related talks
20mSecuring generative AI: Privacy and compliance considerations
50mSecurity as Part of Responsible AI: At Home or At Odds?
35mRisks and Mitigations for a Safe and Responsible AI
01h 16mLifting the Hood on Big AI - The Future of Transparency and Accountability in AI
28mWhy Private AI Matters: Data Sovereignty as the Foundation for Trustworthy AI on the Open Web
30mIntroduction to AI Safety. Let's be ready before it's too late!
44mAI, Corporate Responsibility & Democratic Legitimacy
46mTowards Trustworthy AI: The European Approach