Anthropic is building a way to prove which AI model wrote an answer

Anthropic is working on a way to prove which AI model wrote a given answer. The system is called FlashProve. Researchers described it in a blog post published Tuesday. The goal is to let you confirm that the response you got came from the model you think you are using. It is early research, and no lab has plans to deploy it.

What it does

The idea is called provable inference. That means a cryptographic proof, which is essentially a mathematical guarantee, that an output came from a specific model. Think of the signatures that let you confirm an email really came from its sender. FlashProve is meant to hold up against impostors trying to pass one model off as another. That includes the case where the impostor is the lab that built and runs the model.

Why it matters

Today, you mostly have to trust that an AI service is running the model it says it is. The researchers point to a few situations where proof would help:

• A company secretly swapping in a cheaper model to cut costs.
• A company swapping in a more expensive model to inflate its performance results.
• Third-party claims, such as software scanned by a top security model or a business plan approved by one.

As the researchers put it about those claims: "How can a user verify these claims?"

How it would work

Labs would publish a fingerprint of each model they release. That fingerprint can later be matched to specific outputs and checked with math. It does not reveal the model's weights. Weights are the internal numbers that define how a model behaves, and companies like Anthropic and OpenAI keep them private. So labs could share fingerprints without giving up their most valuable asset. The system is also built to be lightweight, so checking could happen on a local device.

What happens next

Anthropic released a detailed technical specification and working code for parts of the system. Both are public on GitHub under an open-source license. The researchers call the post "intermediate progress, and not a completed research agenda." They want others to take the work and improve it.

Research is ongoing, and there are no plans to deploy FlashProve at Anthropic or any other lab. The material does not say whether other labs will adopt it or when. For now, it is a proposal, not a feature you can use.