{"name":"FlashProve","repo":"anthropics/FlashProve","url":"https://midorreal.com/project/anthropics-flashprove","repository":"https://github.com/anthropics/FlashProve","description":"A research prototype for zero-knowledge proofs of AI inference. Not maintained.","facts":{"license":{"value":"Apache-2.0","source":"https://github.com/anthropics/FlashProve","checked":"22 Nov 2026"},"language":{"value":"Lean","source":"https://github.com/anthropics/FlashProve","checked":"22 Nov 2026"},"releases":{"value":"0","source":"https://github.com/anthropics/FlashProve/releases","checked":"20 Nov 2026"},"contributors":{"value":"1","source":"https://github.com/anthropics/FlashProve/graphs/contributors","checked":"20 Nov 2026"},"archived":{"value":"no","source":"https://github.com/anthropics/FlashProve","checked":"22 Nov 2026"}},"stars":1088.0,"writeup":{"what_it_is":"FlashProve is a research artifact for zero-knowledge proofs of AI inference. It lets an AI provider prove which model answered a request without revealing the model's weights, and anyone can check the proof on a CPU using cp_verify. The project is not maintained and is not accepting contributions.","audience":"Researchers and engineers interested in zero-knowledge proofs of AI inference","claims":[{"kind":"broad","claim":"Lets a provider prove which model answered a request without revealing the weights","excerpt":"lets an AI provider prove which model answered a request, without revealing the model's","status":"too_broad"},{"kind":"specific","claim":"Runs Llama 4 Maverick at full size with a 1,024-token context window on a single NVIDIA H100","excerpt":"FlashProve runs Llama 4 Maverick at full size, with a 1,024-token context window, on a single NVIDIA","status":"not_checked"},{"kind":"specific","claim":"Prover reaches about 110 tokens/second of prefill at 24 requests per batch","excerpt":"The prover reaches about 110 tokens/second of prefill at 24 requests per batch","status":"not_checked"},{"kind":"specific","claim":"Protocol soundness is proved in Lean 4, with a cheating provider caught except with probability about q/2^124 + 2^-101","excerpt":"The protocol's soundness is proved in Lean 4.","status":"not_checked"},{"kind":"specific","claim":"Verification runs on an ordinary CPU in seconds with cp_verify, without the weights","excerpt":"Anyone can check the proof on an ordinary CPU in seconds with `cp_verify`, without the","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-04"},"why_now":null,"owner_supplied":null,"score":{"verdict":"too_early","hype":91,"reality":8,"gap":84,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":{"estimate":true,"lean":"too_close","gap_low":null,"gap_high":null,"rule":"early-2.0"},"changes":[{"date":"2026-10-30T19:29:59+00:00","text":"Repository created","source":"https://github.com/anthropics/FlashProve"}],"cite":"FlashProve is a GitHub project at https://github.com/anthropics/FlashProve written mainly in Lean, described there as \"A research prototype for zero-knowledge proofs of AI inference. Not maintained\".","attribution":"Data from Mid or Real, https://midorreal.com/project/anthropics-flashprove, checked 22 November 2026. Attribution with a link is required.","still_real":null}