Anthropic is working on a system to prove AI models’ identity

Anthropic is developing an open-source system called FlashProve that provides cryptographic proof of which AI model generated a given output. The system, described in a blog post, aims to prevent impostors from disguising one model as another, including by the lab that built and runs it. It could stop AI companies from secretly swapping in cheaper or more expensive models. FlashProve requires labs to publish a model fingerprint that can be matched to outputs and verified cryptographically without revealing the model's weights. The system is designed to be lightweight for local verification. Researchers have released a technical specification and working code on GitHub under an open-source license. The research is ongoing with no immediate plans for deployment, and the post is intended to share intermediate progress for others to improve upon.

(Source：TechCrunch)