Anthropic open-sources FlashProve, its provable inference prototype for MoE models

Anthropic's Frontier Red Team, working with EPFL, MIT and StarkWare, released FlashProve, an open-source prototype that generates a cryptographic proof that a specific language model produced a specific reply. The prover, two verifiers, the formal proofs behind them and a protocol specification are published on GitHub under the Apache 2.0 license, addressing the inability of API users to confirm which model answered a hosted prompt.