Anthropic has published research on FlashProve, a cryptographic method for verifying AI responses using zkSNARK. Claude accelerated proof generation to 110 tokens per second for a model with 400 billion parameters.

FlashProve outlines a way to confirm that a specific response was indeed generated by the stated AI model, without revealing the model's internal parameters. This approach combines privacy and verifiability through zero-knowledge proofs and can be used to authenticate the origin of generated content.

StarkWare researcher Alessandro Chiesa contributed to the development. The zkSNARK technology also underpins Starknet scaling, highlighting the connection between this approach and verification methods already used in blockchain infrastructure.

Anthropic has not announced any direct partnership with Starknet and has not reported using the STRK token.