⬢github Lean · 935 ★ +110 since we first saw it · pushed 6 d ago · Apache-2.0
anthropics/FlashProve
A research prototype for zero-knowledge proofs of AI inference. Not maintained.
FlashProve is a research prototype for zero-knowledge proofs of AI inference: an AI provider can cryptographically prove that a reply came from a committed model without revealing its weights. It runs Llama 4 Maverick on one H100 GPU, produces proofs during inference, and anyone can verify them on a CPU in seconds. The protocol's soundness is formally proved in Lean 4.
Why now: Trending as a most-starred new repo, likely due to the novel combination of zk-proofs with LLM inference and a formally verified soundness theorem in Lean, despite being an unmaintained research artifact.
Who it is for: Cryptography and AI researchers, and agent-driven workflows verifying AI provider claims about which model served a response.
Stars over our 39 snapshots: 825 to 935, since 1 d ago.
Where people talked about it
- ⬢github new repos, most starred 12 min ago
- ⬢github new repos, most starred 18 h ago
API: https://socialmediatrends-api.osmike.com/v1/repos/anthropics/FlashProve