Anthropic open-sources FlashProve, which can verify on an ordinary CPU which model a response came from
Anthropic has open-sourced FlashProve, enabling servers to mathematically prove that a given response came from a specified model, with verification completable on an ordinary CPU. On Meta's Llama 4 Maverick (400B parameters), a single Nvidia H100 generates proofs at 110 tokens/second, each about 19MB, and an ordinary CPU verifies one in about 17 seconds; the only previously known system capable of proving this model, VerInf, generated 93.6GB proofs, with verification taking nearly a day. The system was built by Claude, and because the code is difficult to audit, its security guarantees are mathematically proven in Lean 4.
Why read this
AI score 72
The original text provides comparison figures for proof size and verification time, so readers can judge the feasibility of verifiable inference in deployment.
Topics
View original postNew 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://t.co/OIZWPb6QZa 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://t.co/7ky3Ir8V06 We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (https://t.co/pjWKLHWXyz) and Ben Mayer (@transmissions11).
View original
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://t.co/OIZWPb6QZa 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://t.co/7ky3Ir8V06 We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (https://t.co/pjWKLHWXyz) and Ben Mayer (@transmissions11).
Body
Anthropic just open-sourced FlashProve, which lets you check which AI model actually wrote a reply
checkable on an ordinary CPU.
e.g. if you pay for a top AI model through a reseller, you can’t check whether it quietly served a cheaper one. Anthropic’s FlashProve lets the server prove mathematically which model answered.
Anthropic tested it on Meta’s Llama 4 Maverick, a large 400B-parameter open model, where a single Nvidia H100 GPU produced proofs at 110 tokens/second, each about 19MB.
A regular CPU checks each proof in about 17 seconds, while VerInf, the only earlier system known to prove this model, made 93.6GB proofs that took almost a day to check.
Claude built the system, and because its code is hard to audit, the security guarantees are mathematically proved in Lean 4, a proof-checking language,