Anthropic is working on a system to prove AI models’ identity

| Source: TechCrunch AI

Tags: Anthropic, FlashProve, provable inference, model fingerprinting, cryptographic verification, open source

Anthropic published FlashProve, an open-source research system for cryptographically proving which model produced an output, aimed at stopping silent model swaps even by the lab serving the model.

Details

Anthropic described FlashProve in a blog post on Tuesday. The goal is provable inference: a cryptographic proof that a given output came from a specific model, comparable to email signatures that confirm a sender. Labs would publish a fingerprint for each released model. Outputs can later be matched against it, and the fingerprint does not reveal model weights, so closed-model vendors could take part. Verification is designed to be light enough to run on a local device. The researchers point to two uses: catching a provider that quietly swaps in a cheaper or more expensive model, and checking third-party claims that a task was handled by a particular frontier model, such as a security scan. Anthropic released a technical specification and working code on GitHub under an open-source license. The work is intermediate research, and there are no plans to deploy it at Anthropic or any other lab yet.