Beyond Software: Hardware-Based Security
summary
Argues that replacing a trusted curator with hardware and protocol layers yields practical privacy: the shuffle model amplifies local DP into stronger central DP; Google-style federated learning shows MPC-like aggregation at scale and local DP is currently practical only for companies with >1B users; Intel SGX enclaves (and similar secure elements) let servers run sealed code and keys but suffer reboot and side‑channel issues, solvable with monotonic counters or enclave collisions; Fuchsia’s cobble prototypes such deployments.
AI-written summary · about
chapters (8)
- 0:00Start
- 2:01Phil Rogaway The Moral Character of Cryptographic Work (2015)
- 3:56Local Differential Privacy
- 4:47MPC + Differential Privacy
- 13:28Mixnet (Chaum, C. ACM 1981)
- 16:31Intel Software Guard Extensions (SGX)
- 21:25Remote Attestation
- 26:12Client-Server Architecture
related talks
01h 01mToward Provably Private Federated Learning
20mThe Transformational Power of Confidential Computing
36mEverybody is talking about Confidential Computing - the minimum Devs should know
48mBetrusted: Better Security Through Physical Partitioning
20mKeynote: Privacy at the Hardware Level: Why the Stack Has to Change
19mHow Can We Have Private Access to Intelligence?
13mSurvey of Privacy Solutions for Minimizing MEV
21mSecuring Tomorrow’s Hardware: Dual-State Models in the AI Era