{"id":298630,"date":"2026-02-23T20:44:10","date_gmt":"2026-02-23T20:44:10","guid":{"rendered":"https:\/\/www.newsbeep.com\/nz\/298630\/"},"modified":"2026-02-23T20:44:10","modified_gmt":"2026-02-23T20:44:10","slug":"chinese-ai-companies-distilled-claude-to-improve-own-models-anthropic-says","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/nz\/298630\/","title":{"rendered":"Chinese AI companies &#8216;distilled&#8217; Claude to improve own models, Anthropic says"},"content":{"rendered":"\n<p class=\"yf-vbsvxt\">Feb 23 (Reuters) &#8211; Three Chinese artificial intelligence companies used Claude to improperly obtain capabilities to improve their own models, the chatbot&#8217;s creator Anthropic said \u200cin a blog post on Monday while also making a case for export \u200ccontrols on chips.<\/p>\n<p class=\"yf-vbsvxt\">The announcement follows a memo by <a href=\"https:\/\/www.yahoo.com\/organizations\/openai\/\" data-ylk=\"slk:OpenAI;elm:context_link;itc:0;sec:content-canvas\" class=\"link \" rel=\"nofollow noopener\" target=\"_blank\">OpenAI<\/a> earlier this month, when the startup warned U.S. \u200blawmakers that Chinese AI firm DeepSeek is targeting the ChatGPT maker and the nation&#8217;s leading AI companies to replicate models and use them for its own training.<\/p>\n<p class=\"yf-vbsvxt\">DeepSeek, Moonshot and MiniMax created more than 16 million interactions with Claude using roughly 24,000 fake accounts, in \u200cviolation of Anthropic&#8217;s terms of \u2060service and regional access restrictions, the company said.<\/p>\n<p class=\"yf-vbsvxt\">They used a technique called &#8220;distillation,&#8221; which involves training a less capable model on the outputs \u2060of a stronger one, Anthropic said.<\/p>\n<p class=\"yf-vbsvxt\">&#8220;These campaigns are growing in intensity and sophistication. The window to act is narrow, and the threat extends beyond any single company or region.&#8221;<\/p>\n<p class=\"yf-vbsvxt\">Anthropic warned that \u200billicitly \u200bdistilled models lacked necessary safeguards, creating significant national \u200bsecurity risks. If these models are \u200copen-sourced, the risk multiplies as capabilities spread freely beyond any single government&#8217;s control.<\/p>\n<p class=\"yf-vbsvxt\">Anthropic, which raised $30 billion in its latest funding round and is now valued at $380 billion, said that distillation attacks support the case for export controls: chip access restrictions reduce both direct model training capabilities and the extent of improper distillation.<\/p>\n<p class=\"yf-vbsvxt\">DeepSeek&#8217;s operation targeted reasoning capabilities \u200cacross diverse tasks and the creation of censorship-safe \u200balternatives to policy-sensitive queries, while Moonshot aimed at agentic \u200breasoning and tool use, as well \u200bas coding and data analysis, Anthropic said.<\/p>\n<p class=\"yf-vbsvxt\">MiniMax targeted agentic coding, tool \u200cuse and orchestration and Anthropic detected \u200bthe campaign while it \u200bwas still active \u2014 before MiniMax released the model it was training.<\/p>\n<p class=\"yf-vbsvxt\">&#8220;When we released a new model during MiniMax&#8217;s active campaign, they pivoted within 24 hours, redirecting \u200bnearly half their traffic to \u200ccapture capabilities from our latest system,&#8221; the blog post said.<\/p>\n<p class=\"yf-vbsvxt\">DeepSeek, Moonshot and \u200bMiniMax did not immediately respond to requests for comment.<\/p>\n<p class=\"yf-vbsvxt\">(Reporting by Juby Babu \u200bin Mexico City; Editing by Alan Barona)<\/p>\n","protected":false},"excerpt":{"rendered":"Feb 23 (Reuters) &#8211; Three Chinese artificial intelligence companies used Claude to improperly obtain capabilities to improve their&hellip;\n","protected":false},"author":2,"featured_media":298631,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[365,6166,363,57418,364,7705,5289,111,139,69,164326,145],"class_list":["post-298630","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-anthropic","tag-artificial-intelligence","tag-artificial-intelligence-companies","tag-artificialintelligence","tag-claude","tag-models","tag-new-zealand","tag-newzealand","tag-nz","tag-reasoning-capabilities","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/298630","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/comments?post=298630"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/298630\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media\/298631"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media?parent=298630"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/categories?post=298630"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/tags?post=298630"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}