AI Tech / news
Anthropic Develops FlashProve to Verify Which AI Model Produced an Output
Anthropic is building an open source system called FlashProve that would let anyone confirm an AI response genuinely came from the model it claims to be from, using cryptographic proofs. The company detailed the work in a blog post published Tuesday.
The system is designed to deliver what the researchers term "provable inference" — a cryptographic proof, effectively a mathematical guarantee, that a specific output originated from a specific model. The aim is to let users confirm the answer they received came from the model they believe they are using.
Anthropic compares the mechanism to cryptographic email signatures, which allow recipients to verify that a message truly came from its stated sender. FlashProve is intended to resist impostors trying to pass one model off as another, including the lab that built and operates the model.
According to the researchers, the system could stop an AI company from quietly substituting a cheaper model to reduce costs, or a more costly one to make its performance results look stronger.
FlashProve could also be used to check third-party claims that a particular task was handled by an advanced model. The researchers give the example of a developer asserting that its software library was scanned for vulnerabilities by a frontier cybersecurity model, or a company saying its business plan was approved by a suitable model — and ask how a user could verify such claims.