Sunscreen: FHE secure processing framework for AI and Web3
Thread199 words
Highlights Sunscreen's Secure Processing Framework, which lets teams run LLM inference or smart contract logic on encrypted data via FHE without code rewrites. Its Parasol compiler is claimed to beat Zama by over 2x on some programs.
Key claims
Sunscreen benchmarks beat Zama by over 2x on some programs
SPF is chain-agnostic and needs no rewrite to integrate FHE
Parasol uses a highly parallelized TFHE variant
Amid the recent @a1research__ report on programmable privacy we did, a few projects stood out to me that I will be highlighting in the next few days and weeks.
One of them is @SunscreenTech, which provides an end-to-end platform for secure data coordination and computation with Fully Homomorphic Encryption (FHE).
Their new Secure Processing Framework (SPF) is a completely abstracted platform that enables AI and Web3 teams to train LLMs or run smart contract logic, directly on high-value, proprietary, encrypted data.
Imagine being able to get LLMs to deliver inference or blockchains to execute txs using your private data, without actually seeing it.
SPF is chain-agnostic, and does not require developers to rewrite their existing code to integrate FHE. Its Parasol compiler uses a unique, highly parallelized architecture inspired by transformers in LLMs and pioneers a new variant of the battle-tested TFHE scheme.
Sunscreen's latest benchmarks show how it beats its competitors like @zama by over 2x on some programs.
According to the team, Sunscreen is on a mission to unlock the value of all private data, and I'm excited to continue to follow their journey.
DYOR anon.
In the meantime, check out the full A1 report here:
a1research.io/blog/programmable-privacy-the-next-multi-billion-dollar-infrastructure-layer
1 image on X