Cryptography

Anthropic open-sources proofs that identify which AI model answered

FlashProve can verify inference from committed model weights without revealing them, but the system remains an expensive research prototype.