{"id":788288,"date":"2026-10-01T07:08:11","date_gmt":"2026-10-01T07:08:11","guid":{"rendered":"https:\/\/www.newsbeep.com\/uk\/788288\/"},"modified":"2026-10-01T07:08:11","modified_gmt":"2026-10-01T07:08:11","slug":"irony-alert-openai-whines-that-chinese-model-stole-its-special-ip-that-it-stole-from-everybody-else","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/uk\/788288\/","title":{"rendered":"Irony alert: OpenAI whines that Chinese model stole its special IP that it stole from everybody else"},"content":{"rendered":"<p>                    <a data-tag=\"security\" href=\"https:\/\/www.theregister.com\/tag\/security\" class=\"lab-article-section-link\" rel=\"nofollow noopener\" target=\"_blank\"><\/p>\n<p style=\"\" class=\" lab-article-section lab-article-section-top \">\n                    security\n                <\/p>\n<p>                    <\/a><\/p>\n<p class=\"subtitle below\" style=\"\">\n        US model makers can train on web data &#8211; but distilling theirs is a &#8216;national security risk&#8217;\n    <\/p>\n<p>OpenAI, which hoovered up vast amounts of internet content amid copyright fights, has accused individuals associated with China\u2019s Moonshot AI of being involved in a &#8220;distillation attack&#8221; that began July 1. The house of Altman warns that extracting its models\u2019 reasoning at scale could help rivals train capable models without preserving the same guardrails.<\/p>\n<p>Model distillation is a machine learning technique that can involve using one model\u2019s outputs to train another \u2013 in adversarial cases, by sending bulk queries designed to reproduce the larger model\u2019s reasoning and capabilities.<\/p>\n<p>Both the <a href=\"https:\/\/www.theregister.com\/ai-and-ml\/2026\/09\/09\/us-claims-chinese-ai-companies-core-ai-strategy-is-distilling-american-models\/5295171\" rel=\"nofollow noopener\" target=\"_blank\">feds<\/a> and major US AI companies, including <a href=\"https:\/\/www.theregister.com\/software\/2026\/02\/14\/how-ai-could-eat-itself-using-llms-to-distill-rivals\/4567856\" rel=\"nofollow noopener\" target=\"_blank\">Google<\/a> and <a href=\"https:\/\/www.theregister.com\/software\/2026\/02\/24\/anthropic-misanthropic-toward-chinas-ai-labs\/4119678\" rel=\"nofollow noopener\" target=\"_blank\">Anthropic<\/a>,\u00a0have accused Chinese rivals &#8211; and specifically Moonshot AI &#8211; of using distillation to reproduce capabilities from American models. In a Wednesday blog,\u00a0<a href=\"https:\/\/openai.com\/index\/disrupting-a-coordinated-model-distillation-campaign\/\" rel=\"nofollow noopener\" target=\"_blank\">OpenAI chimed in<\/a>, saying it spotted and ultimately disrupted an adversarial distillation campaign that ran nearly all of July.<\/p>\n<p>\u201cThe operators did not break our encryption, compromise a database, or gain direct access to stored user conversations,\u201d according to the blog. \u201cInstead, they manipulated model interactions so that protected reasoning could be reproduced in forms visible to the requester in a coordinated, scaled manner that violated our terms of service.\u201d<\/p>\n<p>The queries began on July 1, and while they started slowly, \u201cwe observed high-volume spikes on July 24 and 25 consisting of 16,000 requests using a relevant extraction pattern from over 4,000 users,\u201d OpenAI said.<\/p>\n<p>Upon investigating the incident, the AI giant identified related \u201cprompt-pattern activity\u201d across more than 15,000 users. OpenAI fully disrupted the campaign on July 28, we\u2019re told.<\/p>\n<p>While OpenAI said that it&#8217;s unclear whether all of the operators during the July time period were linked to just one rival AI company, the \u201ccore cluster\u201d of the theft came from Moonshot AI, which developed Kimi.<\/p>\n<p>The Register reached out to Moonshot AI for comment and did not receive an immediate response. We also asked OpenAI which of its models were targeted during the July campaign, but did not hear back.<\/p>\n<p>It\u2019s worth noting that, in late July, US President Donald Trump\u2019s Assistant for Science and Technology Michael Kratsios also <a href=\"https:\/\/www.theregister.com\/ai-and-ml\/2026\/07\/23\/senior-white-house-official-claims-chinas-k3-model-stolen-from-anthropic\/5276804\" rel=\"nofollow noopener\" target=\"_blank\">accused Moonshot AI of creating its Kimi K3<\/a> model by distilling Anthropic\u2019s Fable.\u00a0<\/p>\n<p>Anthropic\u2019s <a href=\"https:\/\/www.theregister.com\/ai-and-ml\/2026\/09\/23\/frontier-ai-keeps-racing-despite-calls-to-slow-down\/5298448\" rel=\"nofollow noopener\" target=\"_blank\">Claude Opus 5.5 model<\/a>, released a week ago, comes with a defense against distillation called &#8220;preserved thinking&#8221; that it introduced with Fable 5.1.<\/p>\n<p>\u201cAdversarial distillation poses safety and national security risks,\u201d OpenAI said on Wednesday, echoing earlier gripes from American companies and government officials.\u00a0<\/p>\n<p>\u201cExtracted reasoning could be used to train another model without preserving the safeguards applied to the original model\u2019s user-facing outputs,\u201d OpenAI added. \u201cAt scale, distillation can also accelerate the transfer of advanced capabilities without requiring the same investment in safety. These concerns become heightened as models gain capabilities in dual use domains.\u201d<\/p>\n<p>In response, OpenAI said it banned the model-copying accounts tightened signup and infrastructure controls and expanded monitoring efforts.\u00a0<\/p>\n<p>It also \u201cclosed a pathway that allowed someone who already possessed another user&#8217;s encrypted reasoning to replay it and recover its contents,\u201d and worked with service providers to ensure that this type of distillation activity didn\u2019t just move to third-party services.<\/p>\n<p>Additionally, OpenAI shared the details of its investigation with other AI firms, through the Frontier Model Forum, and government information-sharing programs.\u00ae<\/p>\n","protected":false},"excerpt":{"rendered":"security US model makers can train on web data &#8211; but distilling theirs is a &#8216;national security risk&#8217;&hellip;\n","protected":false},"author":2,"featured_media":788289,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[554,733,4308,86,56,54,55],"class_list":["post-788288","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-technology","tag-uk","tag-united-kingdom","tag-unitedkingdom"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/788288","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/comments?post=788288"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/788288\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media\/788289"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media?parent=788288"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/categories?post=788288"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/tags?post=788288"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}