Anthropic Accuses Moonshot Over Concealed Claude Routing Arabian Post

2026-09-11 04:13:59

(MENAFN- The Arabian Post) clearfix">Anthropic has accused Moonshot AI of secretly sending customer requests meant for its Kimi chatbot to Anthropic's Claude models, alleging that the Beijing-based company concealed the practice from users and presented Claude-generated answers as its own.

The US artificial intelligence developer said on Thursday that it identified one ten-day period in which Moonshot routed almost 300,000 requests through a network of 5,380 fraudulent accounts, with much of the traffic directed to Claude Opus. Anthropic said the activity formed part of a broader effort to gain access to technology that is not directly available from inside China.

Anthropic said it attributed more than 23 million Claude exchanges to Moonshot between May and July. It alleged that the company used intermediaries and large numbers of accounts to evade regional restrictions and detection systems, allowing requests submitted to Kimi to be answered by Claude without users being told that another provider was processing them.

<p>The allegations were included in Anthropic's latest threat intelligence report, which covered activity identified and disrupted between December 2025 and August 2026. The company also accused several other China-based AI laboratories of attempting to extract capabilities from Claude through unauthorised distillation or by routing live traffic to its models.</p><p>Distillation is a standard AI training technique in which a smaller or less capable model learns from outputs produced by a stronger system. Anthropic said the technique becomes improper when competitors obtain those outputs in violation of access restrictions or service terms to reproduce capabilities more cheaply and quickly.</p><p>Moonshot did not issue an immediate public response to the latest allegations. The company, founded in Beijing and best known for its Kimi family of models, has emerged as one of China's most prominent generative AI developers and has been expanding the reach of its products as competition intensifies among domestic and international model providers.</p>See also  Judge overturns Pentagon blacklist against Anthropic <p>Anthropic had already named Moonshot in February when it disclosed what it described as industrial-scale distillation campaigns involving Moonshot, DeepSeek and MiniMax. At that time, Anthropic said the three laboratories generated more than 16 million exchanges through about 24,000 fraudulent accounts. Moonshot alone was linked to more than 3.4 million exchanges in that earlier disclosure.</p><p>The latest findings go further by alleging that some live user requests were redirected directly to Claude rather than merely being used to generate training data. That distinction raises questions about disclosure, data handling and whether customers understood which company's systems were receiving information contained in their prompts.</p><p>Anthropic said some routed requests exposed sensitive user information. Its wider report described instances in which credentials and other potentially confidential material appeared in traffic handled through unauthorised access arrangements, although it did not say that every Moonshot request contained sensitive data.</p><p>The company said its systems blocked or disrupted the activity and that it has strengthened detection measures against coordinated account creation, unusual payment patterns and high-volume access intended to mask the identity of the true user. Anthropic restricts access to Claude from certain regions, including mainland China, and requires customers to comply with its terms of service.</p><p>Anthropic said the broader pattern involved seven China-based laboratories and relied on what it called transfer stations: intermediary services that purchased or controlled Claude access and then relayed traffic on behalf of restricted users. The company said such structures could obscure the origin of requests, distribute activity across many accounts and make coordinated use appear to be customer traffic, complicating enforcement of regional contractual controls.</p>See also  Google raises Pixel prices as AI takes centre stage <p>The dispute also reflects a widening struggle over access to frontier AI models as governments and companies seek to protect intellectual property, computing advantages and model behaviour. US agencies this week separately accused several China-based AI companies of improper efforts to obtain capabilities from leading American systems, allegations rejected by China's Commerce Ministry as groundless.</p><div class='code-block code-block-1'><h2></h2></div>

MENAFN11092026000152002308ID1111651095

Legal Disclaimer:
MENAFN provides the information “as is” without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the provider above.

Anthropic Accuses Moonshot Over Concealed Claude Routing Arabian Post

More Story