Anthropic has introduced FlashProve, an open-source prototype that verifies which AI model generated a response. The system could strengthen trust without exposing commercially sensitive model weights.