Anthropic open-sources FlashProve to cryptographically verify which AI model produced an output

Anthropic researchers have developed FlashProve, an open-source system that issues a cryptographic proof showing which AI model produced a given output. The system, described in a blog post published Tuesday, is designed to let users confirm that a response came from the model they intended to use and to resist impersonation even by the lab that built and runs the model. It could prevent an AI provider from quietly substituting a cheaper model to reduce costs or a more expensive one to exaggerate performance, and it could verify third-party claims that tasks such as vulnerability scans or business plan approvals were handled by advanced models. To use it, labs must publish a model fingerprint that can be matched to outputs and checked cryptographically; the fingerprint does not expose model weights, so the approach can work with closed commercial models. Verification is designed to be lightweight enough to run on a local device. The researchers published a technical specification and working code on GitHub under an open-source license, but said the work is intermediate and that there are no plans to deploy FlashProve at Anthropic or any other lab.