Anthropic Develops FlashProve System to Verify AI Model Identities

On November 4, according to TechCrunch, Anthropic is developing an open-source system called FlashProve, which aims to use cryptographic proofs to verify whether an AI output comes from a specified model.

The system requires model providers to publish model "fingerprints" for users to match and verify against outputs, without leaking model weights. Researchers stated that this can be used to prevent providers from secretly swapping models, and to verify third-party claims about using specific models.

Researchers have publicly released technical specifications and partial system code, but the research is still ongoing. Currently, there are no plans to deploy the system at Anthropic or other laboratories.