Anthropic develops open-source system to cryptographically verify AI model origins

Anthropic has introduced FlashProve, an open-source system designed to provide provable inference—a mathematical guarantee that a specific AI model generated a given output. This mechanism functions similarly to digital signatures, allowing users to confirm the source of an AI-generated response and defending against model impersonation, even if the imposter is the model's creator. The technology could prevent AI firms from secretly substituting cheaper or more expensive models to manipulate cost structures or performance metrics. Furthermore, it enables third parties to verify claims, such as a developer stating their software was scanned by a frontier cybersecurity model. To operate, the system requires AI labs to publish a model fingerprint, which can be cryptographically matched to outputs without disclosing the model's proprietary weights. Researchers released detailed technical specifications and working code for FlashProve on GitHub under an open-source license. Anthropic clarified that this initial release represents intermediate progress, encouraging others to examine and enhance the research, and confirmed there are no immediate plans for deployment at Anthropic or other labs.