--- title: Anthropic found a way to prove which AI model answered... description: Anthropic's Frontier Red Team released FlashProve, an open-source cryptographic system that verifies which AI model generated a response without exposing its... canonical: https://daily.dev/posts/anthropic-found-a-way-to-prove-which-ai-model-answered-you-but-there-s-a-costly-catch-2pxnso4df twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Anthropic found a way to prove which AI model answered you — but there’s a costly catch | daily.dev og:description: Anthropic's Frontier Red Team released FlashProve, an open-source cryptographic system that verifies which AI model generated a response without exposing its... og:url: https://daily.dev/posts/anthropic-found-a-way-to-prove-which-ai-model-answered-you-but-there-s-a-costly-catch-2pxnso4df og:image: https://api.daily.dev/og/posts/2PXnsO4dF.png og:image:alt: Anthropic found a way to prove which AI model answered you — but there’s a costly catch og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Anthropic found a way to prove which AI model answered you — but there’s a costly catch **[The New Stack](https://daily.dev/sources/newstack)** · 14 min read · 0 upvotes · 0 comments ## Summary Anthropic's Frontier Red Team released FlashProve, an open-source cryptographic system that verifies which AI model generated a response without exposing its weights. It uses zkSNARKs and was tested on Meta's Llama 4 Maverick, a 400-billion-parameter model, rather than Claude. On a single Nvidia H100 GPU it processed about 110 prompt tokens per second, producing proofs around 18.78 MB that took 21 seconds to verify on one CPU thread. Generating proofs needs 70 to 95 times more GPU time than standard vLLM inference. It's faster than the earlier VerInf system, which took 14.3 hours and produced a 93.6 GB proof for 1,000 tokens. FlashProve is capped at 1,024 tokens of context and isn't part of the Claude API. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is FlashProve and how does it verify which AI model generated a response? FlashProve is an open-source cryptographic system from Anthropic's Frontier Red Team that uses zkSNARKs to prove a response came from a specific set of model weights without revealing those weights. It was tested on Meta's Llama 4 Maverick, a 400-billion-parameter mixture-of-experts model, rather than Claude, and is released as an unmaintained research artifact on GitHub. _Developers evaluating model-verification tooling can follow zkSNARK-based AI research like this on daily.dev._ ### How much slower is zkSNARK proof generation compared to normal LLM inference? Generating a FlashProve proof needs 70 to 95 times more GPU time than standard inference using vLLM. On a single Nvidia H100 GPU it processed roughly 110 prompt tokens per second while generating proofs, with each proof around 18.78 MB and taking about 21 seconds to verify on a single CPU thread, making it impractical for routine API traffic. _Anyone weighing the overhead of cryptographic verification for AI inference can track benchmarks like these on daily.dev._ ### Why isn't FlashProve available in the Claude API yet? FlashProve remains a research prototype limited to 1,024 tokens of context, and it uses integer-based inference with greedy decoding instead of the floating-point calculations and sampling production models rely on. Anthropic describes the GitHub release as a research artifact that isn't maintained or accepting contributions, with no indication of if or when it will reach the Claude API. _Developers tracking whether verifiable inference reaches production APIs can follow updates on daily.dev._ ## Similar posts on daily.dev - [The Rise of The Theorem Prover](https://daily.dev/posts/the-rise-of-the-theorem-prover-sc6qo4m8q) · Medium · 0 upvotes · 0 comments - [I ran a 9B model trained on Claude's reasoning on my 8GB GPU, and it made me rethink paying for Claude](https://daily.dev/posts/i-ran-a-9b-model-trained-on-claude-s-reasoning-on-my-8gb-gpu-and-it-made-me-rethink-paying-for-clau-mld4wjmej) · XDA Developers · 37 upvotes · 9 comments - [Raising machine-checked security benchmarks to advance hash-based SNARKs through agentic collaboration](https://daily.dev/posts/raising-machine-checked-security-benchmarks-to-advance-hash-based-snarks-through-agentic-collaborati-rlwwj6cre) · Ethereum · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#anthropic](https://daily.dev/tags/anthropic), [#cryptography](https://daily.dev/tags/cryptography), [#ai-security](https://daily.dev/tags/ai-security) [View this post on daily.dev](https://daily.dev/posts/anthropic-found-a-way-to-prove-which-ai-model-answered-you-but-there-s-a-costly-catch-2pxnso4df) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Anthropic found a way to prove which AI model answered you — but there’s a costly catch","url":"https://daily.dev/posts/anthropic-found-a-way-to-prove-which-ai-model-answered-you-but-there-s-a-costly-catch-2pxnso4df","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/anthropic-found-a-way-to-prove-which-ai-model-answered-you-but-there-s-a-costly-catch-2pxnso4df"},"datePublished":"2026-11-04T22:53:50.118Z","dateModified":"2026-11-05T14:46:18.326Z","description":"Anthropic's Frontier Red Team released FlashProve, an open-source cryptographic system that verifies which AI model generated a response without exposing its...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b6c8c3146e5c16020d2d59da9c1ae077?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b6c8c3146e5c16020d2d59da9c1ae077?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/anthropic-found-a-way-to-prove-which-ai-model-answered-you-but-there-s-a-costly-catch-2pxnso4df","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,anthropic,cryptography,ai-security","timeRequired":"PT14M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Anthropic found a way to prove which AI model answered you — but there’s a costly catch"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/anthropic-found-a-way-to-prove-which-ai-model-answered-you-but-there-s-a-costly-catch-2pxnso4df#faq","mainEntity":[{"@type":"Question","name":"What is FlashProve and how does it verify which AI model generated a response?","acceptedAnswer":{"@type":"Answer","text":"FlashProve is an open-source cryptographic system from Anthropic's Frontier Red Team that uses zkSNARKs to prove a response came from a specific set of model weights without revealing those weights. It was tested on Meta's Llama 4 Maverick, a 400-billion-parameter mixture-of-experts model, rather than Claude, and is released as an unmaintained research artifact on GitHub. Developers evaluating model-verification tooling can follow zkSNARK-based AI research like this on daily.dev."}},{"@type":"Question","name":"How much slower is zkSNARK proof generation compared to normal LLM inference?","acceptedAnswer":{"@type":"Answer","text":"Generating a FlashProve proof needs 70 to 95 times more GPU time than standard inference using vLLM. On a single Nvidia H100 GPU it processed roughly 110 prompt tokens per second while generating proofs, with each proof around 18.78 MB and taking about 21 seconds to verify on a single CPU thread, making it impractical for routine API traffic. Anyone weighing the overhead of cryptographic verification for AI inference can track benchmarks like these on daily.dev."}},{"@type":"Question","name":"Why isn't FlashProve available in the Claude API yet?","acceptedAnswer":{"@type":"Answer","text":"FlashProve remains a research prototype limited to 1,024 tokens of context, and it uses integer-based inference with greedy decoding instead of the floating-point calculations and sampling production models rely on. Anthropic describes the GitHub release as a research artifact that isn't maintained or accepting contributions, with no indication of if or when it will reach the Claude API. Developers tracking whether verifiable inference reaches production APIs can follow updates on daily.dev."}}]} ```