Anthropic is developing an open-source system called FlashProve that could let people verify which AI model produced a particular response. The system uses “provable inference”: a cryptographic proof that an output came from a specific model. It is intended to prevent one model from being passed off as another, including cases where a provider secretly […]
