FlashProve, an open-source tool from Anthropic, lets users verify which AI model actually generated a response. The tool combats deceptive practices where companies might secretly swap a more advanced model for a cheaper alternative without disclosure.
The verification process runs locally on the user's machine, ensuring privacy and control. For FlashProve to work, AI labs must publish a cryptographic fingerprint—essentially a digital identity—for each model. This fingerprint serves as proof of origin, making it nearly impossible to misrepresent which system produced a given output.
The tool is available on GitHub at the repository "anthropics/FlashProve," making it freely accessible to developers and organizations that want to implement model verification. By open-sourcing the technology, Anthropic is pushing the industry toward greater transparency in AI deployment and helping users maintain trust in the models they interact with.