Provably Secure AI for Health Data – Powered by Confidential Computing
summary
Claims that Privatemode AI keeps prompts verifiably end-to-end encrypted so infrastructure, cloud and model providers cannot read patient data, using CPU and GPU confidential computing and remote attestation for cryptographic proofs. Implementation runs as a cloud backend plus a local proxy that encrypts prompts, supports Nvidia confidential GPUs and open-source LLMs, and is in production with IBM for Germany’s ePA covering 50 million patients; EY architecture review and a 10‑person‑day pen test report no clear‑text leaks.
AI-written summary · about
chapters (11)
- 0:00Secure AI intro for health data
- 1:27AI challenges with health data
- 2:51What confidential computing enables
- 4:18Edgeless Systems' role
- 5:42Healthcare CC standards
- 6:24Private Mode AI services
- 8:08End-to-end encryption in Private Mode
- 9:52Verifying security via audits
- 10:33Use cases with IBM & Google Cloud
- 11:59Feature roadmap & future
- 12:41Q&A
related talks
14mFaster AI model training with confidential computing
19mPrivate AI: Building Trust Through Verifiable Computation
24mStrictly Confidential: Use Cases & Demos of AI Workflows in TEEs
24mAI, Health & Confidentiality
15mKeynote: Trustworthy AI with Confidential Computing
47mTowards Confidential AI for the Masses!
19mHow Can We Have Private Access to Intelligence?
44mGPU-less, Trust-less, Limit-less: Reimagining the Confidential AI Cloud