Anthropic has unveiled FlashProve, an open-source cryptographic system designed to verify the origin of AI model output.

The library is published under the Apache license. It allows parties to cryptographically prove an AI response originated from specific model weights, without exposing those proprietary weights.