How Can We Have Private Access to Intelligence?
summary
Argues that practical private AI comes from deploying secure hardware (Intel TDX/AMD SEV and NVIDIA confidential computing) plus verifiable supply chain checks and transparency logs, not waiting for fully homomorphic encryption to be fast. Describes remote attestation tied to a public six store ledger and open-source build artifacts, reports confidential inference serving billions of tokens daily with minimal GPU overhead, and shows private chat/inference apps and private training working end to end.
AI-written summary · about
chapters (13)
- 0:00Private AI access intro
- 1:03Always-on AI agents privacy risks
- 2:09Privacy protects your essence
- 3:34AI power concentration democracy threat
- 3:55Local AI models limitations
- 5:18Legal contracts data redaction limits
- 6:05Local privacy with cloud benefits
- 6:27Fully homomorphic encryption challenges
- 8:15Secure hardware trusted execution
- 11:50Secure enclaves for private AI
- 13:14Remote attestation integrity
- 15:02Auditability transparency mechanisms
- 16:47User trust future of private AI
related talks
36mSecurity for AI and Multi-Party Collaboration with Confidential Computing (Oh, and...
37mConfidential AI: Enabling Trustworthy Ecosystems in the Age of Agentic Intelligence - Session
44mGPU-less, Trust-less, Limit-less: Reimagining the Confidential AI Cloud
20mTrusted Hardware for AI Workloads: Extending Confidential...
28mGenerative AI + Proprietary Data
24mPanel Discussion: Proof in Production: Customer Lessons from Confidential AI
15mKeynote: Trustworthy AI with Confidential Computing
47mTowards Confidential AI for the Masses!