Anthropic Open-Sources FlashProve for Verifiable AI Inference
Anthropic introduced and open-sourced FlashProve, a research prototype that uses cryptographic proofs to help verify whether a response came from a specified AI model without exposing the model’s weights. In Anthropic’s published tests on Llama 4 Maverick, each 1,024-token proof was about 19 MB and could be verified in roughly 17 seconds on a 24-thread CPU. The system still requires provider participation, incurs substantial proof-generation overhead, and does not yet provide complete formal guarantees for executable-code equivalence and batched zero-knowledge behavior.
FlashProve offers a cryptographic approach to checking model identity, output tampering, and consistency between tested and deployed models, which could support AI audits and customer trust if providers adopt it.
New Anthropic research: progress on provable inference for large language models.
Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so.
These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers.
For our full write-up, see here: https://www.anthropic.com/research/provable-inference
Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: https://github.com/anthropics/FlashProve
We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (http://rdalal.github.io) and Ben Mayer (@transmissions11).