Anthropic is working on an open source system called FlashProve. It provides a cryptographic proof that an AI response came from a specific model. The system prevents labs from secretly swapping in cheaper or more expensive models. It also lets third parties verify claims about model use in tasks. Labs must publish a model fingerprint for verification without revealing model weights. The system is designed to run on local devices. Anthropic released a technical specification and working code under an open source license. Research is ongoing with no current deployment plans. Anthropic invites others to review and improve the system.