Anthropic develops open source system to cryptographically verify artificial intelligence models

Anthropic announced in a blog post on Tuesday that it is developing an open source system designed to verify which artificial intelligence model produced a specific output, TechCrunch reported. The system, named FlashProve, provides what researchers term provable inference. This cryptographic proof offers a mathematical guarantee that a given output originated from a specific model. [1]

TechCrunch reported that FlashProve functions similarly to cryptographic email signatures, allowing recipients to confirm that a message came from its claimed sender. The system aims to protect users against impostors attempting to disguise one model as another. It enables users to confirm that the response they received genuinely came from the model they intended to use. [1]

As FTMQ Creators reported earlier, software development platform Replit ended its partnership with Anthropic in October citing high costs and model performance capabilities, choosing instead to focus on an agreement with OpenAI. Large language models and reasoning models developed by OpenAI and Anthropic have generated research level mathematical proofs since the mid 2020s. [2][4]

Anthropic operates under a corporate structure that includes an independent Long Term Benefit Trust and utilizes Constitutional AI to train its Claude models using natural language principles. The company has published FlashProve as an open source initiative to establish verifiable model identity across the industry. [1][3]