[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-post-\u002Ffr\u002Fnews\u002Fanthropic-flashprove-prouver-quel-modele-ia-a-repondu":3,"news-related-fr-\u002Ffr\u002Fnews\u002Fanthropic-flashprove-prouver-quel-modele-ia-a-repondu":158,"hreflang-anthropic-flashprove-provable-inference-2026-11":175},{"doc":4,"redirect":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":9,"category":138,"date":139,"description":140,"extension":141,"image":142,"meta":143,"navigation":144,"path":145,"seo":146,"seoDescription":147,"seoTitle":148,"stem":149,"tags":150,"translationKey":156,"updatedAt":8,"__hash__":157},"news_fr\u002Ffr\u002Fnews\u002Fanthropic-flashprove-prouver-quel-modele-ia-a-repondu.md","FlashProve : Anthropic veut prouver quel modèle d'IA vous a vraiment répondu","Paul Lefizelier",null,{"type":10,"value":11,"toc":130},"minimark",[12,31,36,44,61,78,82,93,99,103,119,123],[13,14,15,16,24,25,30],"p",{},"Quand vous payez un modèle précis via une API, vous ne voyez que les tokens renvoyés. Rien ne prouve que le fournisseur ne vous a pas discrètement basculé vers un modèle moins cher. Le 2 novembre 2026, la Frontier Red Team d'Anthropic a ",[17,18,23],"a",{"href":19,"rel":20,"target":22},"https:\u002F\u002Fwww.anthropic.com\u002Fresearch\u002Fprovable-inference",[21],"nofollow","_blank","publié ses avancées sur l'« inférence prouvable »"," et sorti ",[17,26,29],{"href":27,"rel":28,"target":22},"https:\u002F\u002Fgithub.com\u002Fanthropics\u002FFlashProve",[21],"FlashProve",", un prototype open source censé combler ce manque.",[32,33,35],"h2",{"id":34},"comment-ça-marche","Comment ça marche",[13,37,38,39,43],{},"FlashProve fonctionne en deux phases (",[17,40,42],{"href":19,"rel":41,"target":22},[21],"Anthropic",") :",[45,46,47,55],"ul",{},[48,49,50,54],"li",{},[51,52,53],"strong",{},"Hors ligne :"," le fournisseur publie un engagement sur les poids du modèle, une « courte empreinte » qui fixe un jeu de poids précis tout en le gardant secret.",[48,56,57,60],{},[51,58,59],{},"En ligne :"," pour chaque requête, le fournisseur produit un fichier de preuve montrant que le modèle engagé produit bien ces tokens de sortie.",[13,62,63,64,68,69,73,74,77],{},"Toute personne disposant de l'engagement, du prompt et de la réponse peut vérifier la preuve sans les poids. Le ",[17,65,67],{"href":27,"rel":66,"target":22},[21],"dépôt GitHub"," indique que la vérification tourne « sur un CPU ordinaire en quelques secondes » avec l'outil ",[70,71,72],"code",{},"cp_verify",". Le système repose sur des preuves succinctes à divulgation nulle de connaissance (zkSNARK). Il cache les poids, mais pas l'architecture du modèle (",[17,75,42],{"href":19,"rel":76,"target":22},[21],").",[32,79,81],{"id":80},"les-chiffres","Les chiffres",[13,83,84,85,88,89,77],{},"Anthropic l'a testé sur Llama 4 Maverick, un modèle mixture-of-experts de 400 milliards de paramètres, avec ses poids publics en 4 bits. Un seul GPU H100 génère des preuves à environ 110 tokens par seconde, pour une preuve d'environ 19 Mo par requête de 1 024 tokens. Cela représente environ 70 à 95 fois le temps GPU nécessaire pour faire tourner le modèle lui-même sur vLLM (",[17,86,42],{"href":19,"rel":87,"target":22},[21],"). Le dépôt indique une fenêtre de contexte de 1 024 tokens pour l'instant (",[17,90,92],{"href":27,"rel":91,"target":22},[21],"GitHub",[13,94,95,96,77],{},"Anthropic explique que Claude a apporté des améliorations au système de preuve, et que l'équipe a compensé des choix d'optimisation difficiles à auditer en énonçant les garanties de sécurité en Lean, un langage de vérification formelle. Le code, les preuves Lean et la documentation sont publiés sous licence Apache 2.0 (",[17,97,42],{"href":19,"rel":98,"target":22},[21],[32,100,102],{"id":101},"ce-que-ce-nest-pas","Ce que ce n'est pas",[13,104,105,106,109,110,113,114,77],{},"Ce n'est pas un produit. Anthropic parle de « progrès intermédiaires, pas d'un programme de recherche achevé » (",[17,107,42],{"href":19,"rel":108,"target":22},[21],"), le dépôt est présenté comme un artefact de recherche non maintenu (",[17,111,92],{"href":27,"rel":112,"target":22},[21],"), et TechCrunch rapporte qu'aucun déploiement n'est prévu chez Anthropic ni dans un autre laboratoire (",[17,115,118],{"href":116,"rel":117,"target":22},"https:\u002F\u002Ftechcrunch.com\u002F2026\u002F11\u002F03\u002Fanthropic-is-working-on-a-system-to-prove-ai-models-identity\u002F",[21],"TechCrunch",[32,120,122],{"id":121},"pourquoi-cest-important-pour-les-utilisateurs-dia","Pourquoi c'est important pour les utilisateurs d'IA",[13,124,125,126,129],{},"Selon TechCrunch, les chercheurs y voient un moyen d'empêcher un fournisseur de remplacer en secret un modèle par un moins cher, ou par un plus puissant pour gonfler ses résultats de benchmark (",[17,127,118],{"href":116,"rel":128,"target":22},[21],"). Pour ceux qui achètent l'accès aux modèles via des revendeurs ou des routeurs, ou qui paient pour une affirmation du type « vérifié par un modèle de pointe », une preuve vérifiable remplacerait la confiance par un contrôle. Le surcoût de calcul de 70 à 95 fois et la fenêtre de contexte courte montrent que l'usage quotidien est encore loin.",{"title":131,"searchDepth":132,"depth":132,"links":133},"",2,[134,135,136,137],{"id":34,"depth":132,"text":35},{"id":80,"depth":132,"text":81},{"id":101,"depth":132,"text":102},{"id":121,"depth":132,"text":122},"ai","2026-11-04T22:15:00","Le 2 novembre 2026, la Frontier Red Team d'Anthropic a publié FlashProve, un prototype open source qui permet à un fournisseur de prouver qu'une réponse vient d'un modèle donné sans révéler ses poids. C'est du code de recherche, sans déploiement prévu.","md","\u002Fimages\u002Fnews\u002Fanthropic-flashprove-provable-inference.jpg",{},true,"\u002Ffr\u002Fnews\u002Fanthropic-flashprove-prouver-quel-modele-ia-a-repondu",{"title":6,"description":140},"FlashProve d'Anthropic utilise des preuves à divulgation nulle pour vérifier qu'une réponse vient d'un modèle engagé, sans ses poids. Llama 4 Maverick, un H100, environ 110 tokens\u002Fs.","FlashProve : la preuve open source d'Anthropic sur le modèle utilisé","fr\u002Fnews\u002Fanthropic-flashprove-prouver-quel-modele-ia-a-repondu",[151,152,153,154,155],"anthropic","flashprove","cryptographie","securite-ia","open-source","anthropic-flashprove-provable-inference-2026-11","E6H-X-2XihlGR4KfSx6xYbfPFfhzn-kp88r-Qu75y6A",[159,164,170],{"title":160,"description":161,"path":162,"date":139,"image":163,"category":138},"Crouton Alpha : un modèle anonyme, gratuit et à 1 million de tokens arrive sur OpenRouter","OpenRouter a ajouté Crouton Alpha le 3 novembre 2026, un modèle furtif gratuit pour les agents de code, l'appel d'outils et l'analyse de documents, avec un contexte de 1 048 576 tokens. Son développeur est anonyme et les prompts peuvent être conservés.","\u002Ffr\u002Fnews\u002Fopenrouter-crouton-alpha-modele-furtif-gratuit-1m-contexte","\u002Fimages\u002Fnews\u002Fopenrouter-crouton-alpha-stealth-model.jpg",{"title":165,"description":166,"path":167,"date":168,"image":169,"category":138},"Pour l'avocat général de la CJUE, les chatbots comme Gemini peuvent violer les droits des éditeurs de presse","Le 3 novembre 2026, l'avocat général Maciej Szpunar a estimé devant la Cour de justice de l'UE qu'un chatbot qui réutilise des articles dans ses réponses peut porter atteinte aux droits des éditeurs, alors que l'entraînement reste en principe autorisé.","\u002Ffr\u002Fnews\u002Favocat-general-cjue-gemini-droits-editeurs-presse","2026-11-04T17:30:00","\u002Fimages\u002Fnews\u002Fcjeu-gemini-press-publishers-rights.jpg",{"title":171,"description":172,"path":173,"date":168,"image":174,"category":138},"OpenAI relève son bug bounty : jusqu'à 172 000 dollars pour les pires failles de ChatGPT","Le 2 novembre 2026, OpenAI a augmenté les récompenses de son Security Bug Bounty. Une faille critique dans ChatGPT, Codex ou l'API rapporte jusqu'à 30 000 dollars, et une découverte exceptionnelle jusqu'à 172 000 dollars avec les nouveaux bonus.","\u002Ffr\u002Fnews\u002Fopenai-bug-bounty-172000-dollars-chatgpt-codex","\u002Fimages\u002Fnews\u002Fopenai-bug-bounty-172000.jpg",{"en":176,"fr":298,"es":8},{"id":177,"title":178,"author":7,"author_avatar":8,"body":179,"category":138,"date":139,"description":287,"extension":141,"image":142,"meta":288,"navigation":144,"path":289,"seo":290,"seoDescription":291,"seoTitle":292,"stem":293,"tags":294,"translationKey":156,"updatedAt":8,"__hash__":297},"news_en\u002Fnews\u002Fanthropic-flashprove-provable-inference-which-model-answered.md","Anthropic's FlashProve aims to prove which AI model actually answered you",{"type":10,"value":180,"toc":281},[181,193,197,204,218,232,236,246,252,256,270,274],[13,182,183,184,188,189,192],{},"When you pay for a given model through an API, you see only the tokens that come back. Nothing proves the provider did not quietly route you to a cheaper model. On 2 November 2026, Anthropic's Frontier Red Team ",[17,185,187],{"href":19,"rel":186,"target":22},[21],"published progress on \"provable inference\""," and released ",[17,190,29],{"href":27,"rel":191,"target":22},[21],", an open-source prototype meant to close that gap.",[32,194,196],{"id":195},"how-it-works","How it works",[13,198,199,200,203],{},"FlashProve has two phases (",[17,201,42],{"href":19,"rel":202,"target":22},[21],"):",[45,205,206,212],{},[48,207,208,211],{},[51,209,210],{},"Offline:"," the provider publishes a commitment to the model weights, \"a short fingerprint\" that pins down one exact set of weights while keeping them hidden.",[48,213,214,217],{},[51,215,216],{},"Online:"," for each request, the provider produces a proof file showing that the committed model produces those output tokens.",[13,219,220,221,225,226,228,229,77],{},"Anyone holding the commitment, the prompt and the reply can check the proof without the weights. The ",[17,222,224],{"href":27,"rel":223,"target":22},[21],"GitHub repository"," says verification runs \"on an ordinary CPU in seconds\" with a tool called ",[70,227,72],{},". The system relies on zero-knowledge succinct proofs (zkSNARKs). It hides the weights, but hiding the model's architecture is not a goal (",[17,230,42],{"href":19,"rel":231,"target":22},[21],[32,233,235],{"id":234},"the-numbers","The numbers",[13,237,238,239,242,243,77],{},"Anthropic tested it on Llama 4 Maverick, a 400B-parameter mixture-of-experts model, using its public 4-bit weights. A single H100 GPU generates proofs at around 110 tokens per second, with a proof of about 19 MB per 1,024-token request. That is roughly 70 to 95 times the GPU time of running the model itself on vLLM (",[17,240,42],{"href":19,"rel":241,"target":22},[21],"). The repository lists a 1,024-token context window for now (",[17,244,92],{"href":27,"rel":245,"target":22},[21],[13,247,248,249,77],{},"Anthropic says Claude produced improvements to the proof system, and that the team countered hard-to-audit optimisation choices by stating the security guarantees in Lean, a formal verification language. Code, Lean proofs and documentation are released under Apache 2.0 (",[17,250,42],{"href":19,"rel":251,"target":22},[21],[32,253,255],{"id":254},"what-it-is-not","What it is not",[13,257,258,259,262,263,266,267,77],{},"This is not a product. Anthropic calls the post \"intermediate progress, and not a completed research agenda\" (",[17,260,42],{"href":19,"rel":261,"target":22},[21],"), the repository is marked as a research artifact that is not maintained (",[17,264,92],{"href":27,"rel":265,"target":22},[21],"), and TechCrunch reports there are no plans to deploy it at Anthropic or any other lab (",[17,268,118],{"href":116,"rel":269,"target":22},[21],[32,271,273],{"id":272},"why-it-matters-for-ai-users","Why it matters for AI users",[13,275,276,277,280],{},"TechCrunch notes the researchers see it as a way to stop a provider from secretly swapping in a cheaper model, or a stronger one to inflate benchmark results (",[17,278,118],{"href":116,"rel":279,"target":22},[21],"). For anyone buying model access through resellers or routers, or paying for a \"frontier model reviewed this\" claim, a checkable proof would replace trust with verification. The 70 to 95x compute overhead and the short context window show it is still far from everyday use.",{"title":131,"searchDepth":132,"depth":132,"links":282},[283,284,285,286],{"id":195,"depth":132,"text":196},{"id":234,"depth":132,"text":235},{"id":254,"depth":132,"text":255},{"id":272,"depth":132,"text":273},"On 2 November 2026 Anthropic's Frontier Red Team released FlashProve, an open-source prototype that lets a provider prove a reply came from a committed model without revealing its weights. It is research code, with no deployment planned.",{},"\u002Fnews\u002Fanthropic-flashprove-provable-inference-which-model-answered",{"title":178,"description":287},"Anthropic's FlashProve uses zero-knowledge proofs so anyone can check a reply came from a committed model, without the weights. Llama 4 Maverick, one H100, about 110 tokens\u002Fs.","FlashProve: Anthropic's open-source proof of which model answered","news\u002Fanthropic-flashprove-provable-inference-which-model-answered",[151,152,295,296,155],"cryptography","ai-safety","DZ0HRWbeipwSLGA0spCdcHlIJHZrwUjIev3EWhjOeUY",{"id":5,"title":6,"author":7,"author_avatar":8,"body":299,"category":138,"date":139,"description":140,"extension":141,"image":142,"meta":377,"navigation":144,"path":145,"seo":378,"seoDescription":147,"seoTitle":148,"stem":149,"tags":379,"translationKey":156,"updatedAt":8,"__hash__":157},{"type":10,"value":300,"toc":371},[301,309,311,316,326,336,338,346,351,353,364,366],[13,302,15,303,24,306,30],{},[17,304,23],{"href":19,"rel":305,"target":22},[21],[17,307,29],{"href":27,"rel":308,"target":22},[21],[32,310,35],{"id":34},[13,312,38,313,43],{},[17,314,42],{"href":19,"rel":315,"target":22},[21],[45,317,318,322],{},[48,319,320,54],{},[51,321,53],{},[48,323,324,60],{},[51,325,59],{},[13,327,63,328,68,331,73,333,77],{},[17,329,67],{"href":27,"rel":330,"target":22},[21],[70,332,72],{},[17,334,42],{"href":19,"rel":335,"target":22},[21],[32,337,81],{"id":80},[13,339,84,340,88,343,77],{},[17,341,42],{"href":19,"rel":342,"target":22},[21],[17,344,92],{"href":27,"rel":345,"target":22},[21],[13,347,95,348,77],{},[17,349,42],{"href":19,"rel":350,"target":22},[21],[32,352,102],{"id":101},[13,354,105,355,109,358,113,361,77],{},[17,356,42],{"href":19,"rel":357,"target":22},[21],[17,359,92],{"href":27,"rel":360,"target":22},[21],[17,362,118],{"href":116,"rel":363,"target":22},[21],[32,365,122],{"id":121},[13,367,125,368,129],{},[17,369,118],{"href":116,"rel":370,"target":22},[21],{"title":131,"searchDepth":132,"depth":132,"links":372},[373,374,375,376],{"id":34,"depth":132,"text":35},{"id":80,"depth":132,"text":81},{"id":101,"depth":132,"text":102},{"id":121,"depth":132,"text":122},{},{"title":6,"description":140},[151,152,153,154,155]]