FlashProve, announced by Anthropic, uses cryptographic proofs to verify which model generated an output.
Companies publish a fingerprint for each model; it does not expose the model’s internal weights, and verification is designed to run locally.
The code and specification are open source on GitHub.
Research is ongoing, with no immediate deployment plans from Anthropic or other labs.
Why is it important?
Today, users can't know for sure whether the response they're getting is coming from the powerful model they're paying for or from a cheaper, less capable program. FlashProve could prevent companies from secretly replacing models in an attempt to cut costs.
At the same time, the system could be used to verify third-party claims. If, for example, a company claims that its business plan or code was tested by a specific, top-tier model, FlashProve would offer a way to cross-check the validity of that claim.
How it works
For the system to work, companies must publish a digital “footprint” for each of their models. This footprint allows the result to be verified through calculations, without revealing the model’s internal data, thus protecting the companies’ trade secrets.
↳
READ ALSO
HACKS.GR / RELATED
AI Unlocked
Why artificial intelligence centers use aircraft engines for electricity
↗
The system is designed to be lightweight and can be performed even on a local device. Anthropic has made the FlashProve code and technical description available on GitHub as open source. Research remains ongoing and there are no immediate plans for the system to be implemented by the company itself or other labs.
>Stay connected with Hacks.gr
Set us as a preferred source on Google and find us on social media.
Preferred source
↗
end of article
Top comments (0)