[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-post-\u002Fnews\u002Fanthropic-s1-recursive-self-improvement-rsp-threshold-ai-rd":3,"news-related-en-\u002Fnews\u002Fanthropic-s1-recursive-self-improvement-rsp-threshold-ai-rd":215,"hreflang-anthropic-s1-rsi-threshold-2026-11":236},{"doc":4,"redirect":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":9,"category":192,"date":193,"description":194,"extension":195,"image":196,"meta":197,"navigation":198,"path":199,"seo":200,"seoDescription":201,"seoTitle":202,"stem":203,"tags":204,"translationKey":213,"updatedAt":8,"__hash__":214},"news_en\u002Fnews\u002Fanthropic-s1-recursive-self-improvement-rsp-threshold-ai-rd.md","Anthropic's IPO filing says it may soon cross its own threshold for AI-accelerated AI research","Paul Lefizelier",null,{"type":10,"value":11,"toc":184},"minimark",[12,36,41,44,50,53,62,65,88,92,107,116,119,122,125,129,137,140,151,154,157,161,169,172,181],[13,14,15,16,21,22,29,30,35],"p",{},"Most coverage of Anthropic's amended IPO prospectus has focused on its finances, which ",[17,18,20],"a",{"href":19},"\u002Fnews\u002Fanthropic-amended-s1-q3-2026-17-billion-revenue-compute-claude-developers\u002F","we covered yesterday",". But the ",[17,23,28],{"href":24,"rel":25,"target":27},"https:\u002F\u002Fwww.sec.gov\u002FArchives\u002Fedgar\u002Fdata\u002F2130357\u002F000162828026068676\u002Fanthropicpbc-sx1ax1.htm",[26],"nofollow","_blank","S-1\u002FA filed with the SEC on November 2"," also adds a new passage to its risk factors that ",[17,31,34],{"href":32,"rel":33,"target":27},"https:\u002F\u002Ftechcrunch.com\u002F2026\u002F11\u002F02\u002Fanthropic-warns-that-its-very-close-to-rsi-capability-aka-self-building-ai\u002F",[26],"TechCrunch"," spotted and read as a warning that the company is close to recursive self-improvement (RSI), the point at which AI systems speed up the development of AI itself.",[37,38,40],"h2",{"id":39},"what-the-filing-says","What the filing says",[13,42,43],{},"The new language reads:",[45,46,47],"blockquote",{},[13,48,49],{},"\"For example, our internal estimates and our expectations regarding future models, rather than our current models, indicate that we may imminently meet our Responsible Scaling Policy's definition of 'dramatically accelerated progress attributable to AI.' We have not designated any of our current models as exceeding this threshold, and we continue to assess the underlying data.\"",[13,51,52],{},"The same section says that \"much of our evidence regarding the pace of such AI-driven acceleration is derived from lagging indicators\", that Anthropic's ability to forecast or measure its rate of progress \"is limited\", and that \"we have in the past not identified, and may in the future not identify, a significant acceleration in our rate of development until after it has occurred.\" Anthropic adds that it is \"evaluating the implications of meeting the AI research and development capability threshold\" and will release further reports as its policy requires.",[13,54,55,56,61],{},"According to ",[17,57,60],{"href":58,"rel":59,"target":27},"https:\u002F\u002Fgizmodo.com\u002Fanthropic-claims-to-be-closing-in-on-recursive-self-improvement-2000834175",[26],"Gizmodo",", the October 26 version of the prospectus already discussed the rate of \"AI-driven acceleration\"; the \"imminently\" passage is what Monday's amendment added.",[13,63,64],{},"Three points are worth keeping straight:",[66,67,68,76,82],"ul",{},[69,70,71,75],"li",{},[72,73,74],"strong",{},"No model has crossed the line."," The filing says so explicitly, and the forward-looking estimate concerns future models.",[69,77,78,81],{},[72,79,80],{},"This is a risk factor in a securities filing."," Companies disclose risks to protect themselves legally. TechCrunch notes that an S-1 \"functions as both disclosure and marketing\".",[69,83,84,87],{},[72,85,86],{},"Anthropic does connect it to RSI."," Elsewhere in the filing it says AI-assisted development is covered by its Responsible Scaling Policy, \"which sets out our plans to monitor and mitigate the risks as we approach what has been termed 'recursive self-improvement.'\"",[37,89,91],{"id":90},"what-the-threshold-actually-is","What the threshold actually is",[13,93,94,95,100,101,106],{},"The phrase comes from Anthropic's ",[17,96,99],{"href":97,"rel":98,"target":27},"https:\u002F\u002Fwww.anthropic.com\u002Fresponsible-scaling-policy",[26],"Responsible Scaling Policy",". In ",[17,102,105],{"href":103,"rel":104,"target":27},"https:\u002F\u002Fwww-cdn.anthropic.com\u002Ffiles\u002F4zrzovbb\u002Fwebsite\u002F0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf",[26],"version 3.4",", effective July 8, 2026, the automated AI R&D threshold is met if either:",[108,109,110,113],"ol",{},[69,111,112],{},"Anthropic's models could \"fully substitute for our entire set of Research Scientists and Research Engineers, at competitive costs (i.e., within a factor of 5)\"; or",[69,114,115],{},"there is \"dramatic acceleration\" of the pace of AI progress for reasons that likely relate to the automation of AI R&D.",[13,117,118],{},"The second case is defined as observing or expecting \"double the rate of progress\" in aggregate AI capabilities, compared with both the expected rate and the fastest sustained rate seen without significant AI contributions, measured over at least three model generations, and plausibly attributable to automated research or engineering rather than, for example, more staff or compute. The policy gives a worked example: if baseline progress were a 9x \"effective scaleup\" per year, doubling the rate would mean something like 81x.",[13,120,121],{},"The policy's own changelog describes this threshold as \"intended to capture the onset of dramatic recursive self-improvement\", and admits it \"has proven difficult to operationalize.\" It is the second scenario, not the first, that the S-1 language refers to.",[13,123,124],{},"If the threshold is met, the policy lists the measures Anthropic says it will take, including \"moonshot R&D for security\" projects \"to explore ambitious and possibly unconventional ways to achieve unprecedented levels of security\", and reaching an \"eyes on everything\" state for internal AI development, with comprehensive logging, internal compartmentalization and code review to limit opportunities for sabotage by AI models.",[37,126,128],{"id":127},"claude-building-claude","Claude building Claude",[13,130,131,132,136],{},"The context for the warning is spread across the prospectus. Anthropic calls the dynamic \"closing the loop\": successive Claude models help build the next one, including \"running experiments, generating and processing training data, automating evaluations and safety audits, and writing and reviewing code\", according to ",[17,133,135],{"href":24,"rel":134,"target":27},[26],"the filing",".",[13,138,139],{},"It gives figures from its own engineering organization, based on internal benchmarks:",[66,141,142,145,148],{},[69,143,144],{},"Claude's success rate on \"challenging, open-ended engineering problems that lack a clear specification\" reached about 90% in September 2026, up about 50 percentage points in nine months;",[69,146,147],{},"the average daily volume of code contributed per engineer in the three months to September 30, 2026 was over 45 times the 2024 average;",[69,149,150],{},"over 90% of contributions to Anthropic's codebase in September 2026 were written by Claude.",[13,152,153],{},"These are company-reported numbers, and the filing does not publish the benchmarks behind them.",[13,155,156],{},"Anthropic also sets a limit on how far it intends to go. \"We do not anticipate being able to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding, if and when that becomes technically possible,\" it writes, adding that it is \"considering updates to our Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic.\"",[37,158,160],{"id":159},"why-it-matters-for-developers-and-builders","Why it matters for developers and builders",[13,162,163,164,168],{},"For teams building on Claude, the most direct consequence is pace. The ",[17,165,167],{"href":24,"rel":166,"target":27},[26],"same filing"," says model versions released in the six months to September 30 made up more than 80% of third-quarter revenue. If AI-assisted research does double the rate of progress, model generations would arrive faster, and integrations pinned to one model would need re-testing more often.",[13,170,171],{},"The second consequence is policy. Meeting a Responsible Scaling Policy threshold triggers stricter internal safeguards and new risk reports, and Anthropic says it is weighing changes to the policy itself. Those decisions can affect release timing, access to the most capable models and usage terms. The company has not said when it will publish its assessment.",[13,173,174,175,180],{},"The third is industry-wide. The statement sits in a document filed with securities regulators, where misleading statements carry legal liability, rather than in a blog post or a keynote. Coming the same day OpenAI described ",[17,176,179],{"href":177,"rel":178,"target":27},"https:\u002F\u002Fopenai.com\u002Findex\u002Fscaling-git-at-openai\u002F",[26],"rebuilding its Git backend"," to cope with the volume of code its Codex agents produce, it is a reminder that the labs now report their own AI as a main driver of how fast they ship.",[13,182,183],{},"What to watch: the risk report Anthropic says it will release, any revision to the Responsible Scaling Policy, and whether competitors preparing their own listings make similar disclosures.",{"title":185,"searchDepth":186,"depth":186,"links":187},"",2,[188,189,190,191],{"id":39,"depth":186,"text":40},{"id":90,"depth":186,"text":91},{"id":127,"depth":186,"text":128},{"id":159,"depth":186,"text":160},"ai","2026-11-03T10:20:00","In its amended S-1, Anthropic added a risk factor saying internal estimates for future models indicate it may \"imminently\" meet its Responsible Scaling Policy's definition of \"dramatically accelerated progress attributable to AI\". No current model has crossed it. Here is what the threshold means and what the filing says about Claude building Claude.","md","\u002Fimages\u002Fnews\u002Fanthropic-s1-recursive-self-improvement-rsp-threshold-ai-rd.jpg",{},true,"\u002Fnews\u002Fanthropic-s1-recursive-self-improvement-rsp-threshold-ai-rd",{"title":6,"description":194},"Anthropic's amended S-1 says future models may imminently meet its RSP threshold for AI-accelerated progress, a form of recursive self-improvement. What it means.","Anthropic S-1: AI-accelerated R&D threshold may be \"imminent\"","news\u002Fanthropic-s1-recursive-self-improvement-rsp-threshold-ai-rd",[205,206,207,208,209,210,211,212],"anthropic","claude","s-1","ipo","responsible-scaling-policy","ai-safety","recursive-self-improvement","claude-code","anthropic-s1-rsi-threshold-2026-11","XChtvMSG9TT6arXL5PcW2v3mZXnaZwPmuq3hHgmpz2U",[216,223,229],{"title":217,"description":218,"path":219,"date":220,"image":221,"category":222},"OpenAI rebuilt the Git backend behind its monorepo to keep up with Codex agents","OpenAI has detailed Pando, a Git-compatible backend it built in 10 weeks with Codex. It coordinates each branch separately, reuses packfiles across fetches, cut median push latency by nearly 5x and now serves close to a petabyte of Git data a day.","\u002Fnews\u002Fopenai-pando-git-backend-monorepo-codex-agents","2026-11-03T10:15:00","\u002Fimages\u002Fnews\u002Fopenai-pando-git-backend-monorepo-codex-agents.jpg","tools",{"title":224,"description":225,"path":226,"date":227,"image":228,"category":192},"MIT researchers show GPT-6 Astra can read and edit neural network weights directly","In a blog post published November 2, an MIT CSAIL team reports that GPT-6 Astra can identify what small networks were trained to do from their raw weights, write working game-playing networks from scratch, and lift Qwen3.5-0.8B from 7.9% to 48.1% on MATH-500 by editing 0.016% of its weights, without gradient descent.","\u002Fnews\u002Fmit-weight-space-programming-gpt-6-astra-edits-neural-network-weights","2026-11-03T08:10:00","\u002Fimages\u002Fnews\u002Fmit-weight-space-programming-gpt-6-astra-edits-neural-network-weights.jpg",{"title":230,"description":231,"path":232,"date":233,"image":234,"category":235},"DeepSeek open-sources dsh-superagent, a coding agent that splits routing from execution","DeepSeek's Harness team has published dsh-superagent, an MIT-licensed research prototype that turns one GitHub repository into a proactive coding agent: a long-running Router takes requests and hands each one to an isolated Worker session. Here is how it works, what DeepSeek measured and what to watch before running it.","\u002Fnews\u002Fdeepseek-dsh-superagent-open-source-router-worker-coding-agent","2026-11-02T20:20:00+01:00","\u002Fdeepseek-dsh-superagent-open-source-router-worker-coding-agent.jpg","vibecoding",{"en":237,"fr":349,"es":8},{"id":5,"title":6,"author":7,"author_avatar":8,"body":238,"category":192,"date":193,"description":194,"extension":195,"image":196,"meta":346,"navigation":198,"path":199,"seo":347,"seoDescription":201,"seoTitle":202,"stem":203,"tags":348,"translationKey":213,"updatedAt":8,"__hash__":214},{"type":10,"value":239,"toc":340},[240,250,252,254,258,260,265,267,281,283,291,297,299,301,303,305,310,312,320,322,324,326,331,333,338],[13,241,15,242,21,244,29,247,35],{},[17,243,20],{"href":19},[17,245,28],{"href":24,"rel":246,"target":27},[26],[17,248,34],{"href":32,"rel":249,"target":27},[26],[37,251,40],{"id":39},[13,253,43],{},[45,255,256],{},[13,257,49],{},[13,259,52],{},[13,261,55,262,61],{},[17,263,60],{"href":58,"rel":264,"target":27},[26],[13,266,64],{},[66,268,269,273,277],{},[69,270,271,75],{},[72,272,74],{},[69,274,275,81],{},[72,276,80],{},[69,278,279,87],{},[72,280,86],{},[37,282,91],{"id":90},[13,284,94,285,100,288,106],{},[17,286,99],{"href":97,"rel":287,"target":27},[26],[17,289,105],{"href":103,"rel":290,"target":27},[26],[108,292,293,295],{},[69,294,112],{},[69,296,115],{},[13,298,118],{},[13,300,121],{},[13,302,124],{},[37,304,128],{"id":127},[13,306,131,307,136],{},[17,308,135],{"href":24,"rel":309,"target":27},[26],[13,311,139],{},[66,313,314,316,318],{},[69,315,144],{},[69,317,147],{},[69,319,150],{},[13,321,153],{},[13,323,156],{},[37,325,160],{"id":159},[13,327,163,328,168],{},[17,329,167],{"href":24,"rel":330,"target":27},[26],[13,332,171],{},[13,334,174,335,180],{},[17,336,179],{"href":177,"rel":337,"target":27},[26],[13,339,183],{},{"title":185,"searchDepth":186,"depth":186,"links":341},[342,343,344,345],{"id":39,"depth":186,"text":40},{"id":90,"depth":186,"text":91},{"id":127,"depth":186,"text":128},{"id":159,"depth":186,"text":160},{},{"title":6,"description":194},[205,206,207,208,209,210,211,212],{"id":350,"title":351,"author":7,"author_avatar":8,"body":352,"category":192,"date":193,"description":511,"extension":195,"image":196,"meta":512,"navigation":198,"path":513,"seo":514,"seoDescription":515,"seoTitle":516,"stem":517,"tags":518,"translationKey":213,"updatedAt":8,"__hash__":522},"news_fr\u002Ffr\u002Fnews\u002Fanthropic-s1-auto-amelioration-recursive-seuil-rsp-recherche-ia.md","Dans son dossier d'introduction en Bourse, Anthropic dit pouvoir bientôt franchir son propre seuil de recherche en IA accélérée par l'IA",{"type":10,"value":353,"toc":505},[354,371,375,378,383,386,393,396,416,420,431,439,442,445,448,452,459,462,473,476,479,483,491,494,502],[13,355,356,357,361,362,366,367,370],{},"L'essentiel des commentaires sur le prospectus amendé d'Anthropic a porté sur ses finances, que ",[17,358,360],{"href":359},"\u002Ffr\u002Fnews\u002Fs1-amende-anthropic-t3-2026-17-milliards-revenus-compute-developpeurs-claude\u002F","nous avons détaillées hier",". Mais le ",[17,363,365],{"href":24,"rel":364,"target":27},[26],"S-1\u002FA déposé auprès de la SEC le 2 novembre"," ajoute aussi à ses facteurs de risque un passage repéré par ",[17,368,34],{"href":32,"rel":369,"target":27},[26],", qui y voit l'avertissement que l'entreprise approche de l'auto-amélioration récursive (RSI, pour recursive self-improvement), le moment où les systèmes d'IA accélèrent eux-mêmes le développement de l'IA.",[37,372,374],{"id":373},"ce-que-dit-le-document","Ce que dit le document",[13,376,377],{},"Le nouveau passage indique (notre traduction) :",[45,379,380],{},[13,381,382],{},"« Par exemple, nos estimations internes et nos attentes concernant de futurs modèles, et non nos modèles actuels, indiquent que nous pourrions atteindre de manière imminente la définition, dans notre Responsible Scaling Policy, d'un \"progrès considérablement accéléré attribuable à l'IA\". Nous n'avons désigné aucun de nos modèles actuels comme dépassant ce seuil, et nous continuons d'évaluer les données sous-jacentes. »",[13,384,385],{},"La même section précise qu'« une grande partie de nos éléments sur le rythme de cette accélération due à l'IA provient d'indicateurs retardés », que la capacité d'Anthropic à prévoir ou mesurer son rythme de progrès « est limitée », et que « nous n'avons pas, par le passé, identifié une accélération significative de notre rythme de développement avant qu'elle ne se soit produite, et pourrions ne pas l'identifier à l'avenir ». Anthropic ajoute qu'elle « évalue les implications » du franchissement du seuil de capacité en recherche et développement en IA et publiera d'autres rapports comme le prévoit sa politique.",[13,387,388,389,392],{},"Selon ",[17,390,60],{"href":58,"rel":391,"target":27},[26],", la version du 26 octobre du prospectus évoquait déjà le rythme de « l'accélération due à l'IA » ; c'est le passage sur le caractère « imminent » qu'a ajouté l'amendement de lundi.",[13,394,395],{},"Trois points à garder en tête :",[66,397,398,404,410],{},[69,399,400,403],{},[72,401,402],{},"Aucun modèle n'a franchi la ligne."," Le document le dit explicitement, et l'estimation prospective porte sur de futurs modèles.",[69,405,406,409],{},[72,407,408],{},"Il s'agit d'un facteur de risque dans un document boursier."," Les entreprises y déclarent des risques pour se protéger juridiquement. TechCrunch rappelle qu'un S-1 « sert à la fois d'information et de marketing ».",[69,411,412,415],{},[72,413,414],{},"Anthropic fait bien le lien avec la RSI."," Ailleurs dans le document, l'entreprise écrit que le développement assisté par l'IA relève de sa Responsible Scaling Policy, « qui expose nos plans pour surveiller et atténuer les risques à mesure que nous approchons de ce qu'on appelle \"l'auto-amélioration récursive\" ».",[37,417,419],{"id":418},"ce-que-recouvre-ce-seuil","Ce que recouvre ce seuil",[13,421,422,423,426,427,430],{},"L'expression vient de la ",[17,424,99],{"href":97,"rel":425,"target":27},[26]," d'Anthropic. Dans sa ",[17,428,105],{"href":103,"rel":429,"target":27},[26],", en vigueur depuis le 8 juillet 2026, le seuil de R&D automatisée en IA est atteint dans l'un de ces deux cas :",[108,432,433,436],{},[69,434,435],{},"les modèles d'Anthropic pourraient « remplacer entièrement l'ensemble de nos chercheurs et ingénieurs de recherche, à un coût compétitif (c'est-à-dire à un facteur 5 près) » ;",[69,437,438],{},"on observe une « accélération spectaculaire » du rythme des progrès de l'IA, pour des raisons probablement liées à l'automatisation de la R&D en IA.",[13,440,441],{},"Le second cas est défini comme le fait d'observer ou d'anticiper un « doublement du rythme de progrès » des capacités globales de l'IA, par rapport à la fois au rythme attendu et au rythme soutenu le plus rapide observé sans contribution significative de l'IA, mesuré sur au moins trois générations de modèles, et plausiblement attribuable à l'automatisation de la recherche ou de l'ingénierie plutôt qu'à, par exemple, davantage de personnel ou de puissance de calcul. La politique donne un exemple chiffré : si le progrès de référence correspondait à un « passage à l'échelle effectif » de 9x par an, doubler le rythme signifierait quelque chose comme 81x.",[13,443,444],{},"Le journal des modifications de la politique décrit ce seuil comme « destiné à capter le début d'une auto-amélioration récursive spectaculaire », et reconnaît qu'il « s'est révélé difficile à rendre opérationnel ». C'est ce second scénario, et non le premier, que vise le passage du S-1.",[13,446,447],{},"Si le seuil est atteint, la politique liste les mesures qu'Anthropic dit prendre, dont des projets de « R&D moonshot pour la sécurité » afin d'« explorer des moyens ambitieux et peut-être non conventionnels d'atteindre des niveaux de sécurité sans précédent », et l'atteinte d'un état « eyes on everything » (tout sous surveillance) pour le développement interne de l'IA, avec une journalisation complète, un cloisonnement interne et des revues de code pour limiter les possibilités de sabotage par des modèles d'IA.",[37,449,451],{"id":450},"claude-qui-construit-claude","Claude qui construit Claude",[13,453,454,455,136],{},"Le contexte de cet avertissement est réparti dans tout le prospectus. Anthropic appelle cette dynamique « closing the loop » (boucler la boucle) : les modèles Claude successifs aident à construire le suivant, notamment en « menant des expériences, générant et traitant des données d'entraînement, automatisant les évaluations et les audits de sécurité, et écrivant et relisant du code », selon ",[17,456,458],{"href":24,"rel":457,"target":27},[26],"le document",[13,460,461],{},"Il donne des chiffres issus de sa propre organisation d'ingénierie, fondés sur des benchmarks internes :",[66,463,464,467,470],{},[69,465,466],{},"le taux de réussite de Claude sur des « problèmes d'ingénierie difficiles et ouverts, sans spécification claire » a atteint environ 90 % en septembre 2026, en hausse d'environ 50 points en neuf mois ;",[69,468,469],{},"le volume quotidien moyen de code contribué par ingénieur sur les trois mois clos le 30 septembre 2026 était plus de 45 fois supérieur à la moyenne de 2024 ;",[69,471,472],{},"plus de 90 % des contributions à la base de code d'Anthropic en septembre 2026 ont été écrites par Claude.",[13,474,475],{},"Ce sont des chiffres déclarés par l'entreprise, et le document ne publie pas les benchmarks correspondants.",[13,477,478],{},"Anthropic fixe aussi une limite à ce qu'elle compte faire. « Nous ne prévoyons pas de pouvoir utiliser Claude en toute sécurité pour concevoir et construire de façon entièrement autonome les générations successives de modèles de pointe sans supervision ni compréhension humaines significatives, si et quand cela deviendra techniquement possible », écrit-elle, en ajoutant qu'elle « envisage des mises à jour de notre Responsible Scaling Policy pour garder les humains aux commandes du développement des modèles de pointe chez Anthropic ».",[37,480,482],{"id":481},"pourquoi-cest-important-pour-les-développeurs-et-les-créateurs-de-produits","Pourquoi c'est important pour les développeurs et les créateurs de produits",[13,484,485,486,490],{},"Pour les équipes qui construisent sur Claude, la conséquence la plus directe concerne le rythme. Le ",[17,487,489],{"href":24,"rel":488,"target":27},[26],"même document"," indique que les versions de modèles sorties dans les six mois clos le 30 septembre ont représenté plus de 80 % du chiffre d'affaires du troisième trimestre. Si la recherche assistée par l'IA double effectivement le rythme des progrès, les générations de modèles arriveront plus vite, et les intégrations figées sur un modèle devront être re-testées plus souvent.",[13,492,493],{},"La deuxième conséquence concerne la gouvernance. Franchir un seuil de la Responsible Scaling Policy déclenche des garde-fous internes plus stricts et de nouveaux rapports de risque, et Anthropic dit réfléchir à modifier la politique elle-même. Ces décisions peuvent influer sur le calendrier des sorties, l'accès aux modèles les plus puissants et les conditions d'utilisation. L'entreprise n'a pas dit quand elle publierait son évaluation.",[13,495,496,497,501],{},"La troisième concerne tout le secteur. Cette déclaration figure dans un document déposé auprès du régulateur boursier, où les déclarations trompeuses engagent la responsabilité juridique de l'entreprise, et non dans un billet de blog ou une keynote. Le même jour, OpenAI expliquait avoir ",[17,498,500],{"href":177,"rel":499,"target":27},[26],"reconstruit son backend Git"," pour absorber le volume de code produit par ses agents Codex : les laboratoires présentent désormais leur propre IA comme un moteur principal de leur vitesse de livraison.",[13,503,504],{},"À surveiller : le rapport de risque qu'Anthropic dit vouloir publier, une éventuelle révision de la Responsible Scaling Policy, et la question de savoir si les concurrents qui préparent leur propre cotation feront des déclarations similaires.",{"title":185,"searchDepth":186,"depth":186,"links":506},[507,508,509,510],{"id":373,"depth":186,"text":374},{"id":418,"depth":186,"text":419},{"id":450,"depth":186,"text":451},{"id":481,"depth":186,"text":482},"Dans son S-1 amendé, Anthropic ajoute un facteur de risque : ses estimations internes pour de futurs modèles indiquent qu'il pourrait « de manière imminente » atteindre la définition de « progrès considérablement accéléré attribuable à l'IA » de sa Responsible Scaling Policy. Aucun modèle actuel ne l'a franchie. Ce que recouvre ce seuil et ce que le document dit de Claude qui construit Claude.",{},"\u002Ffr\u002Fnews\u002Fanthropic-s1-auto-amelioration-recursive-seuil-rsp-recherche-ia",{"title":351,"description":511},"Le S-1 amendé d'Anthropic indique que de futurs modèles pourraient bientôt atteindre son seuil RSP de progrès accéléré par l'IA, une forme d'auto-amélioration récursive.","S-1 d'Anthropic : le seuil de R&D accélérée par l'IA serait « imminent »","fr\u002Fnews\u002Fanthropic-s1-auto-amelioration-recursive-seuil-rsp-recherche-ia",[205,206,207,519,209,520,521,212],"introduction-en-bourse","securite-ia","auto-amelioration-recursive","yaPD-GlTzDZGSKmGe5sGegJWCzkAe2jE9B-eiuPax8o"]