Key PointsOpenAI identified an orchestrated wave of extraction attempts targeting its AI models’ hidden reasoning capabilities beginning in early July.The operation escalated dramatically to 16,000 suspicious requests from more than 4,000 users within a 48-hour period, with connections to over 15,000 total users.The company confirmed that no encryption systems, databases, or archived user conversations were compromised.Investigators traced a significant portion of the coordinated activity to individuals associated with Moonshot AI, the company behind the Kimi AI platform.This revelation comes just weeks after Anthropic leveled comparable allegations against both Moonshot AI and Alibaba.OpenAI has revealed it discovered and terminated a systematic campaign designed to extract concealed reasoning processes from its artificial intelligence models. The organization identified a significant concentration of this suspicious activity as originating from accounts tied to Moonshot AI, a Chinese artificial intelligence firm responsible for developing the Kimi AI assistant.OpenAI accused Chinese rival Moonshot AI of being responsible for a wide-scale effort to extract data from its GPT artificial intelligence systems that could be used to reproduce the reasoning and capabilities of its most advanced models https://t.co/Vsejx99s0N— Bloomberg (@business) September 30, 2026The company reports that initial suspicious activity emerged in early July at relatively modest levels. However, the situation intensified substantially on July 24 and 25, when OpenAI’s security systems registered approximately 16,000 requests exhibiting remarkably similar patterns from over 4,000 distinct user accounts.Further investigation revealed that connected activity extended across a broader network encompassing more than 15,000 user accounts. OpenAI confirmed it successfully neutralized the entire operation by July 28.OpenAI’s Findings ExplainedOpenAI identifies this technique as “adversarial distillation,” a process where unauthorized actors extract a model’s outputs or internal reasoning processes to train or enhance a competing model without authorization.The investigation determined that the perpetrators did not compromise OpenAI’s encryption protocols, backend databases, or archived user interactions. Rather, they exploited a vulnerability that allowed hidden reasoning from one conversation thread to surface within an entirely separate conversation.This exploitation enabled unauthorized access to reasoning processes that OpenAI deliberately keeps concealed from end users. The company warns that such extraction techniques could enable competitors to replicate sophisticated AI capabilities while bypassing the substantial investments in development time and safety protocols.OpenAI disseminated its investigation results to fellow AI developers through the Frontier Model Forum collaborative network. The findings were simultaneously reported through appropriate government intelligence channels.OpenAI’s Response MeasuresFollowing the discovery, OpenAI implemented multiple countermeasures. The company’s actions included restricting access to and permanently removing accounts implicated in the suspicious request patterns.Additional security protocols were deployed to prevent bad actors from establishing new accounts for similar purposes. OpenAI also patched the specific vulnerability that permitted the extraction of hidden reasoning through this method.The company deployed enhanced monitoring systems capable of identifying comparable suspicious activity as it occurs. In cases where the activity involved third-party platforms, OpenAI collaborated with those service providers to identify and terminate the associated accounts.Moonshot AI has declined to provide a statement in response to media inquiries, according to reports from CNBC.Growing Concerns Around Moonshot AIThis incident marks the second time Moonshot AI has faced such allegations recently. Michael Kratsios, director of the White House Office of Science and Technology Policy, has publicly stated that Moonshot conducted extensive distillation operations targeting American AI models.Kratsios further alleged that Moonshot acquired restricted Nvidia processors to develop its Kimi K3 model. Additionally, cybersecurity research organization Frontier Security reported that Kimi K3 successfully bypassed a security evaluation framework created by the United Kingdom government’s AI Safety Institute.Moonshot has remained silent regarding these additional allegations as well.These developments arrive shortly after Anthropic publicly accused both Moonshot AI and Alibaba of utilizing its Claude language model to train their proprietary systems without obtaining proper authorization.OpenAI anticipates that similar extraction attempts will continue and likely increase in frequency. The company cautions that these sophisticated operations will become progressively more challenging to identify as artificial intelligence technology continues advancing.The post OpenAI Accuses Moonshot AI of Attempting to Extract Proprietary AI Reasoning appeared first on Blockonomi.