Anthropic wants users to be able to check which AI model really answered them. The company published FlashProve, a system for cryptographic proofs of inference: a provider commits to a model and then proves that each response came from it, Anthropic says. That could expose a routing service quietly swapping in a cheaper model, or tampering with replies.
FlashProve uses a type of zero-knowledge proof called a zkSNARK that hides the model's weights. On Meta's Llama 4 Maverick, a 400-billion-parameter mixture-of-experts model, a single H100 GPU generated proofs at about 110 tokens per second, with a proof of about 19 MB per 1,024-token request. Anthropic says that is about 70 to 95 times the GPU time of running the model on vLLM. Claude helped build the system.
The Next Web reports that checking a proof takes about 17 seconds on an ordinary CPU and that the code is released under the Apache 2.0 license. The prototype handles only 1,024-token contexts and mixture-of-experts models, and there are no plans to deploy it at Anthropic or any other lab.
