Report / Tech
Published by NOISE
Anthropic develops open-source system to prove AI model identity
From Evening NOISE.
Anthropic on Tuesday announced FlashProve, an open-source system intended to provide cryptographic proof, called provable inference, that a given output came from a specific AI model. The system requires labs to publish a fingerprint for each model that can later be matched to outputs without revealing model weights, and Anthropic released a technical specification and working code on GitHub, with verification designed to run on a local device. FlashProve is meant to prevent a lab or third party from disguising one model as another, such as by secretly substituting a cheaper model or inflating performance claims, though research is ongoing and Anthropic said it has no plans to deploy the system.
Saved versions
These dates record changes to NOISE’s copy. They are not event dates.