From The Next Web · Ana Maria Constantin · · 1m
Anthropic's FlashProve proves which AI model answered you
Anthropic releases an open-source tool that cryptographically proves an AI reply came from the model a provider claims to have used.
In brief
Anthropic releases an open-source tool that cryptographically proves an AI reply came from the model a provider claims to have used. Anthropic released FlashProve, an open-source prototype using cryptographic proofs to verify that AI replies came from the claimed model, not a cheaper substitute. Originally reported by…
The problem: you cannot verify which model answered
Today users cannot tell whether a router swapped in a cheaper model or an attacker tampered with replies. Anthropic's FlashProve tackles this with provable inference.
How FlashProve works: cryptographic proof of model identity
The provider publishes a fingerprint of its model weights. Each reply comes with a cryptographic proof that those weights produced it. The proof reveals nothing about the weights themselves.
Speed and scale: seventy to ninety five times slower
Anthropic tested FlashProve on Meta's Llama 4 Maverick, a four hundred billion parameter mixture of experts model.
Real world uses and current limits
Anthropic sees uses for independent evaluators, customers of routing services, and regulators checking approved models.
The main idea
Anthropic released FlashProve, an open-source prototype using cryptographic proofs to verify that AI replies came from the claimed model, not a cheaper substitute.
