Security Thoughts on Untrusted Nodes

One of the biggest challenges we will face relative to executing compute over data will be how to execute over encrypted/secure data.

We do not have a current plan of attack other than our current plan of:

1. Using public data (only) to start
2. Encouraging the use of subnets on trusted nodes.

Projects

However, there is quite a bit of thinking in the context of MPC (multi-party compute). Below is a partial list of folks who have already thought about it (in alpha order):

• ABY^3 - a semi-honest implementation of ABY 3 and Fast Database Joins for Secret Shared Data. https://github.com/ladnir/aby3
• Awesome Homomorphic Encryption (HE) - list of resources around HE https://github.com/jonaschn/awesome-he
• Batch DualEx - https://github.com/osu-crypto/batchDualEx
• Crypten - Privacy Preserving Machine Learning built on PyTorch https://github.com/facebookresearch/CrypTen
• EMP - Efficient, multi-party computation toolkit - https://github.com/emp-toolkit/emp-tool
• Honeybadger MPC - HoneyBadgerMPC is a robust MPC-based confidentiality layer for blockchains. https://github.com/initc3/HoneyBadgerMPC
• Nucypher - enables proxy re-encryption without access to the private keys and/or plaintext https://www.nucypher.com/
• TF Encrypted - a framework for encrypted machine learning in TensorFlow https://github.com/tf-encrypted/tf-encrypted

Papers

• Refereed Delegation of Computation - https://www.cs.tau.ac.il/~canetti/CRR11.pdf
• Incentivizing Outsourced Computation - http://netecon.seas.harvard.edu/NetEcon08/Papers/Belenkiy_08.pdf