Anthropic Develops FlashProve for AI Model Identity Verification
Anthropic is creating an open source system called FlashProve to verify which AI model generated a specific output, as detailed in a blog post published on November 3, 2026. This system aims to provide what researchers refer to as “provable inference”—a cryptographic proof that guarantees a given output originated from a specific model.
FlashProve is designed to function similarly to cryptographic email signatures, allowing users to confirm that a response genuinely comes from the model it claims to represent. The system is intended to be resilient against impostors attempting to misrepresent one model as another, even if the impostor is the lab that developed the model.
Researchers highlight that FlashProve could prevent AI companies from secretly substituting a less capable model to reduce costs or using a more advanced model to exaggerate performance metrics. This verification capability could also be useful for validating claims made by third parties regarding the use of advanced models for specific tasks.
To operate effectively, the system requires AI labs to publish a “fingerprint” for each model they release. This fingerprint can be matched to specific outputs and verified cryptographically. Importantly, these fingerprints do not disclose a model’s internal weights, which are proprietary to companies like Anthropic and OpenAI. This allows labs to share fingerprints without compromising their valuable intellectual property.
FlashProve is designed to be lightweight, enabling verification to be conducted on local devices. Alongside the announcement, researchers have released a detailed technical specification and working code for various components of the system, which is available under an open source license on GitHub.
Ongoing research into the system is in progress, and there are currently no plans for deployment at Anthropic or any other lab. Researchers clarify that the purpose of the announcement is to share their intermediate progress and encourage others to build upon their work.
Source: TechCrunch