Anthropic releases FlashProve prototype, using zkSNARK to verify AI responses originate from a specified model without exposing weights
Anthropic releases open-source research prototype FlashProve, using zkSNARK cryptographic proofs to confirm AI responses come from the declared model while not revealing model weights. On a single H100 GPU, generating proofs for the 400-billion-parameter Llama 4 Maverick model yields approximately 110 tokens per second; each proof is roughly 19 MB. Verification on standard CPUs takes about 17 seconds, with overhead ranging from 70x to 95x compared to normal inference.
Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original