Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek
China demands the United States revoke its federal procurement ban against DeepSeek and threatens retaliation over supply-chain security actions. Anthropic's threat report details coordinated distillation campaigns by Chinese labs Moonshot and DeepSeek, accusing them of using thousands of fraudulent accounts and dark web purchases to covertly route user queries to Claude. The captured data reportedly helps Chinese startups build cheaper models, while Beijing has launched an internal inquiry into its domestic AI firms following the allegations.
Listen to Live Briefing
Real-time synthesized voice briefing · Live Feeds Desk
- ✓ China urged the United States to revoke its exclusion order restricting federal procurement of DeepSeek technology.
- ✓ Anthropic accused Chinese startups Moonshot and DeepSeek of covertly routing user requests to Claude to train their own systems.
- ✓ Beijing warned of potential retaliation against United States actions that threaten its interests.
What changed
China officially demanded that Washington rescind its DeepSeek procurement exclusion order while opening an internal probe into its own AI startups.
Live updates
-
China Urges US to Revoke DeepSeek Ban Amid AI Theft Claims
China demands the United States revoke its federal procurement ban against DeepSeek and threatens retaliation over supply-chain security actions. Anthropic's threat report details coordinated distillation campaigns by Chinese labs Moonshot and DeepSeek, accusing them of using thousands of fraudulent accounts and dark web purchases to covertly route user queries to Claude. The captured data reportedly helps Chinese startups build cheaper models, while Beijing has launched an internal inquiry into its domestic AI firms following the allegations.
Why it matters
The dispute centers on allegations that Chinese artificial intelligence companies harvest proprietary American models through API abuse and model distillation. Anthropic states that Chinese labs utilized tens of thousands of fake accounts to extract reasoning data and train domestic systems. This intellectual property clash has triggered federal procurement restrictions in the United States and warnings of retaliatory trade measures from Beijing.
What is confirmed
- China urged the United States to revoke its exclusion order restricting federal procurement of DeepSeek technology.
- Anthropic accused Chinese startups Moonshot and DeepSeek of covertly routing user requests to Claude to train their own systems.
- Beijing warned of potential retaliation against United States actions that threaten its interests.
Still unconfirmed
- Chinese labs bought stolen Claude accounts on the dark web.
- DeepSeek and Moonshot relied on GPT-6 Astra outputs to train Kimi K3.1 Pro and V4.1-Flash 1017.
What to watch next
- Specific retaliatory measures announced by Beijing against United States economic interests
- Results from Beijing's official inquiry into domestic AI companies
confidence 90%Sources used for this update (9)
- gadgetsnow.indiatimes.com — China Challenges US DeepSeek Ban as AI Distillation Dispute Grows
- www.phynnet.com — Anthropic Exposes Chinese AI Firms' LLM Theft Playbook
- ingoaweigel.substack.com — China’s DeepSeek Probe Isn’t About What the Headlines Say
- www.securityweek.com — Stolen AI Reasoning Could Help China Build Cheaper Rivals
- www.phynnet.com — Anthropic Accuses Moonshot, DeepSeek of Secretly Routing ...
- thenationalpulse.com — Top AI Firm Says Chinese Labs Stole U.S. Tech Using 24,000 ...
- darkwebdecoded.com — Dark Web Stolen Accounts Fuel China’s AI Theft
- news.ingage.institute — Hidden AI Theft Risks Exposed - The INGAGE Institute
- www.explainx.ai — DeepSeek, Moonshot Distilled GPT-6 Astra? What Is Verified ...
-
DHS Bans DeepSeek as Anthropic Fights AI Distillation
The Department of Homeland Security banned the Chinese AI model DeepSeek from federal civilian agencies due to national security and data privacy risks. This follows Anthropic's efforts to stop Chinese labs from harvesting reasoning data from Claude. Anthropic previously reported that defenses reduced extracted reasoning tokens by 90% to 99% after seven labs used 24,000 fake accounts to generate 16 million exchanges. While Anthropic continues to offer agentic coding and platform tools, external analysts suggest that combining distillation combat with chip access limits could double the US lead over China within a year.
Why it matters
Distillation occurs when a smaller model is trained using the outputs of a larger, more capable model to mimic its reasoning. This practice allows competitors to bypass the massive compute costs of original training. The US government is now treating these technical vulnerabilities as national security threats.
What is confirmed
- DHS banned DeepSeek from federal civilian agencies citing national security and data privacy risks.
- Anthropic provides a coding agent called Claude Code and a platform for creating AI agents.
Still unconfirmed
- Limiting remote access to US AI chips and combating distillation could double the American AI lead over China within one year.
What to watch next
- Evidence of new distillation methods used by Chinese labs to bypass Anthropic defenses
- Further federal bans on other Chinese AI models like Moonshot AI or MiniMax
confidence 90%Sources used for this update (4)
- claude.com — Claude Code by Anthropic | AI Coding Agent, Terminal, IDE
- platform.claude.com — Anthropic - Claude Platform
- ifp.org — How to Increase America’s AI Lead Over China | IFP
- cyberscoop.com — DHS to ban DeepSeek from federal civilian agencies | CyberScoop
-
Anthropic deploys defenses against Chinese AI distillation campaigns
Anthropic has implemented strengthened defenses to stop Chinese AI labs from harvesting reasoning data from Claude. On October 12, 2026, the company reported that these measures reduced the amount of extracted reasoning tokens by 90% to 99%. The attacks involved at least seven Chinese labs, including DeepSeek, Moonshot AI, and MiniMax, which used 24,000 fraudulent accounts to generate 16 million exchanges. While the data harvest has dropped, Anthropic noted that blocked requests are now billed to clients and the long-term durability of these gains is unknown.
Why it matters
Distillation allows less capable models to be trained on the outputs of stronger ones to replicate their performance. This conflict involves high-stakes competition between US and Chinese AI developers over model capabilities and export controls. Anthropic manages this risk through a tiered model system, including the restricted Mythos and public Fable series.
What is confirmed
- Anthropic detailed expanded defenses against distillation attacks on October 12, 2026.
- DeepSeek, Moonshot AI, and MiniMax used 24,000 fraudulent accounts and 16 million Claude exchanges to conduct distillation attacks.
- New security measures reduced the extraction of reasoning data by 90% to 99%.
- Anthropic released Claude Fable 5 and Claude Mythos 5 in June 2026, with Fable 5 routing distillation-related requests to the less capable Claude Opus.
- Claude Mythos 5.1 and Fable 5.1 were released in September 2026.
Still unconfirmed
- At least seven artificial intelligence labs in China extracted reasoning traces from Claude models.
- Blocked requests resulting from distillation defenses are now billed to clients.
What to watch next
- Evidence regarding the long-term durability of the 90-99% reduction in data extraction.
- Official US government confirmation of sanctions or Entity List actions against Moonshot AI.
confidence 90%Sources used for this update (9)
- builtin.com — What Is Anthropic? | Built In
- www.unite.ai — Anthropic Reports Distillation Defenses Reduced Extracted Reasoning Tokens
- en.wikipedia.org — Claude Mythos - Wikipedia
- www.dplooy.com — Anthropic Exposes Chinese AI Distillation Attacks 2026
- www.el-fondo.com — Anthropic shuts down Chinese AI distillation campaigns
- aiweekly.co — Anthropic Says New Defenses Cut Distillation Harvest 90-99%
- economictimes.indiatimes.com — ET AI Explainer: Why are US AI labs wary of 'distillation ...
- lapaasvoice.com — Anthropic tackles AI distillation attacks on Claude
- www.actuia.com — Anthropic locks Claude against distillation, clients pay
-
Anthropic Accuses Chinese Labs of Massive Data Theft
Anthropic accuses Chinese artificial intelligence developers including Alibaba, Moonshot AI, and DeepSeek of executing industrial-scale distillation campaigns to extract capabilities from Claude. According to Anthropic, these firms utilized roughly 24,000 fraudulent accounts to generate in excess of 16 million exchanges with Claude to train cheaper copycat models. The allegations place Moonshot AI at the center of a escalating technological conflict between the United States and China, drawing scrutiny from American officials who have warned of potential sanctions and Entity List actions against the developer for targeting the Kimi K3 model.
Why it matters
Model distillation allows developers to mine answers from advanced artificial intelligence systems to build cheaper competing models, effectively cutting down development cycles and research costs. This practice has triggered a fierce technological struggle between the United States and foreign AI developers over API security, data protection, and intellectual property. OpenAI previously disrupted a similar coordinated campaign on September 30, 2026, tying the activity directly to individuals associated with Moonshot AI.
What is confirmed
- Anthropic accuses Chinese AI firms including Alibaba, Moonshot AI, and DeepSeek of conducting distillation campaigns to extract capabilities from Claude.
- Anthropic stated that DeepSeek, Moonshot AI, and MiniMax used approximately 24,000 fraudulent accounts to generate over 16 million Claude exchanges.
- Moonshot AI faces United States scrutiny and potential sanctions and Entity List actions following accusations of using industrial-scale distillation to improve its Kimi K3 model.
Still unconfirmed
- Independent reporting stated that the accused firms were contacted for comment regarding the API abuse allegations.
What to watch next
- Official United States government announcements regarding sanctions or Entity List additions for Moonshot AI or other implicated Chinese developers.
- Implementation of stricter API-abuse detection and provenance controls by frontier artificial intelligence model providers.
confidence 90%Sources used for this update (3)
- interestingengineering.com — Anthropic claims Chinese AI firms used Claude to train models
- www.wionews.com — Moonshot AI faces US sanctions over ‘distillation attacks’ to ...
- letsdatascience.com — Anthropic says three Chinese AI labs used fraudulent accounts ...
-
Anthropic and OpenAI Accuse Chinese AI Firms of Distillation
Anthropic and OpenAI have accused Chinese artificial intelligence developers of mining their models' answers to train cheaper copycats through a process known as model distillation. Anthropic previously accused Moonshot AI of secretly using Claude to build its own model. OpenAI subsequently disrupted a coordinated campaign on September 30, 2026, attributing core activity to individuals associated with Moonshot AI. These events place Moonshot AI at the center of a technological conflict between the United States and China.
Why it matters
Model distillation allows developers to extract reasoning capabilities and answers from advanced models to train competing systems. This practice has become a focal point of tension between US artificial intelligence labs and Chinese firms. Anthropic operates as a prominent American public benefit corporation headquartered in San Francisco and is the creator of the Claude large language model series.
What is confirmed
- Anthropic and OpenAI accused Chinese AI developers of mining their models answers to train cheaper copycats.
- OpenAI disrupted a coordinated campaign on September 30, 2026, attributing a core cluster of the activity to individuals associated with Moonshot AI.
- Anthropic was founded in January 2021 by former members of OpenAI, including siblings Daniela Amodei and Dario Amodei.
- Anthropic was valued at US$965 billion in a May 2026 Series H funding round.
What to watch next
- Further disclosures from Anthropic or OpenAI regarding the scope of distillation campaigns by Alibaba, Moonshot AI, and DeepSeek
- Official responses from Chinese AI developers accused of distillation campaigns
confidence 100%Sources used for this update (7)
- www.anthropic.com — Newsroom \ Anthropic
- www.anthropic.com — Home \\ Anthropic
- en.wikipedia.org — Anthropic - Wikipedia
- claude.com — The AI for Problem Solvers | Claude by Anthropic
- en.wikipedia.org — Dario Amodei - Wikipedia
- academy.claude.com — Claude Academy · Learn to work and build with Claude
- www.scientificamerican.com — What is AI model distillation, and why is it so hard to stop? | Scientific American
-
OpenAI and Anthropic Accuse Moonshot AI of Model Distillation
OpenAI has joined Anthropic in publicly accusing users linked to the Chinese artificial intelligence company Moonshot AI of attempting to extract the reasoning capabilities of advanced models through distillation. OpenAI disclosed that it disrupted a coordinated campaign on September 30, 2026, attributing a core cluster of the activity to individuals associated with Moonshot AI. This follows earlier accusations from Anthropic stating that Moonshot AI secretly used Claude to build its own model, placing the firm at the center of a technological conflict between the United States and China.
Why it matters
Model distillation campaigns represent a rising security challenge for American artificial intelligence laboratories seeking to protect proprietary technology from foreign competitors. US security agencies previously reported that multiple China-based entities extracted billions of tokens from major domestic firms using automated queries and fraudulent accounts. The public alignment of OpenAI and Anthropic against Moonshot AI escalates pressure on foreign artificial intelligence developers accused of intellectual property theft.
What is confirmed
- OpenAI disclosed that it disrupted a coordinated campaign to extract the protected reasoning of its models on September 30, 2026.
- OpenAI attributed a core cluster of the distillation activity to individuals associated with Moonshot AI.
- Anthropic stated that Moonshot AI secretly used Claude to build its own model.
What to watch next
- Formal regulatory responses or sanctions from the White House or US security agencies regarding the distillation campaigns.
- Additional disclosures from Google or SpaceX regarding similar data extraction attempts.
- Official statements or defense responses from Moonshot AI addressing the accusations made by OpenAI and Anthropic.
confidence 100%Sources used for this update (3)
- sigmawire.net — Moonshot AI vs Anthropic: The Claude Data Theft Row Explained
- labs.cloudsecurityalliance.org — Reasoning Extraction: Distillation as an Attack on Model ...
- biz.heraldcorp.com — OpenAI joins Anthropic in accusing Moonshot AI of 'distillation'
-
Anthropic Details Distillation Campaigns Involving Chinese Firms
Anthropic published a threat intelligence report detailing industrial-scale model distillation campaigns conducted by Alibaba, Moonshot AI, and DeepSeek. US security agencies including the NSA, FBI, and CISA state that six China-based artificial intelligence companies extracted billions of tokens from American firms including OpenAI, Anthropic, Google, and SpaceX. Anthropic separately accused Moonshot AI of concealing the routing of customer requests from its Kimi chatbot to Claude models. The activity utilized thousands of fraudulent accounts to execute millions of automated queries, prompting investigations by Chinese regulators into domestic data leaks.
Why it matters
These revelations expose vulnerabilities in the security architecture surrounding advanced commercial artificial intelligence models. The involvement of multiple Chinese firms indicates a coordinated, systematic approach to bypassing domestic training bottlenecks by harvesting proprietary outputs from Western developers. Regulatory responses from both US security agencies and the Cyberspace Administration of China highlight the geopolitical friction generated by large-scale model extraction.
What is confirmed
- US intelligence agencies accuse DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI of extracting billions of tokens from OpenAI, Anthropic, Google, and SpaceX at industrial scale.
- Anthropic accused Moonshot AI of secretly sending customer requests meant for its Kimi chatbot to Anthropic's Claude models.
Still unconfirmed
- Anthropic's September 2026 report describes disrupted misuse across seven distinct areas covering cyber operations.
What to watch next
- Formal regulatory penalties or sanctions from US agencies against the named Chinese artificial intelligence firms.
- Official findings and enforcement actions from the Cyberspace Administration of China regarding domestic data leaks and prompt rerouting.
- Additional threat intelligence disclosures from other Western artificial intelligence providers regarding model extraction.
confidence 95%Sources used for this update (3)
- dailysecurityreview.com — NSA, CISA, FBI Accuse Six Chinese AI Firms of Model ...
- thearabianpost.com — Anthropic accuses Moonshot over concealed Claude routing
- postcutoff.com — Anthropic threat intelligence report: AI-orchestrated ...
-
Anthropic Details Chinese AI Distillation Campaigns
Anthropic accuses DeepSeek, Moonshot AI, and MiniMax of large-scale distillation attacks against Claude, operating through thousands of fraudulent accounts to run roughly 190 million queries. US agencies including the NSA, FBI, and CISA state that six China-based AI companies ran industrial-scale extraction campaigns starting at least by late 2024. China's Cyberspace Administration responded by summoning named companies and investigating DeepSeek and Moonshot AI over domestic data leaks and user prompt rerouting.
Why it matters
The dispute centers on competitive technological edge and data security across dual ecosystems. While Washington alleges that Chinese labs systematically harvested capabilities from United States frontier models, Beijing has turned its regulatory focus inward to examine potential data exposure resulting from the process.
What is confirmed
- Anthropic accused DeepSeek, Moonshot AI, and MiniMax of large-scale distillation attacks on Claude.
- An Anthropic threat intelligence report alleges that seven Chinese AI labs used thousands of fraudulent accounts to run roughly 190 million queries.
- The NSA, FBI and CISA stated on September 8 that six China-based AI companies allegedly ran industrial-scale campaigns to extract capabilities from US frontier models dating to at least late 2024.
- China's Cyberspace Administration summoned every company named in Anthropic's report and opened a formal investigation into DeepSeek and Moonshot AI.
Still unconfirmed
- The distillation activity likely occurred with Chinese government awareness according to US agencies, which remains an allegation rather than a judicial finding.
What to watch next
- Findings or penalties from China's Cyberspace Administration probe into DeepSeek and Moonshot AI
- Potential blocking or defensive measures considered by the United States against AI distillation techniques
confidence 100%Sources used for this update (7)
- www.alphamatch.ai — Anthropic Accuses Chinese AI Firms of Data Theft: A Deeper Look
- kimkj.com — [AI Library] Trump, the Diplomacy of Insult – Chapter 5: The DeepSeek AI Shock and the Dual Ecosystems ....
- thehackernews.com — Cybercrime | Breaking Cybersecurity News | The Hacker News
- www.winzheng.com — Anthropic Names Seven Chinese AI Labs: 190 Million Queries ...
- letsdatascience.com — US Agencies Allege Chinese AI Distillation Campaigns
- www.chosun.com — US-China AI Distillation Sparks Data Leak Concerns - 조선일보
- www.ainchina.com — China Probes DeepSeek and Moonshot: How Anthropic's Report ...
-
China probes DeepSeek, Moonshot AI over data for Anthropic
China's Cyberspace Administration is investigating DeepSeek and Moonshot AI over allegations that they rerouted user prompts through Anthropic's Claude systems to train domestic models. The probe focuses on data that may have reached Anthropic, with 190 million extractions previously reported. Alibaba's larger campaign is not the current focus.
Why it matters
The investigation is part of a broader scrutiny of Chinese AI labs and their data practices. Anthropic had published a report detailing the extractions, revealing a specific translation prompt method used to expose Claude's internal deliberation processes. This development highlights the challenges in regulating AI and data use in the industry.
What is confirmed
- China's regulator is investigating DeepSeek and Moonshot over data that may have reached Anthropic.
- Anthropic published a report detailing 190 million extractions.
Still unconfirmed
- Alibaba's larger campaign is not the focus of the current investigation.
What to watch next
- Further details on the investigation's findings
- Any updates on potential penalties for DeepSeek and Moonshot AI
- The impact on China's AI industry and regulations
confidence 90%Sources used for this update (3)
- kimkj.com — [AI서재] 트럼프, 모욕의 외교 - 제5장 딥시크(DeepSeek) AI 충격과 엔비디아·화웨이....
- thenextweb.com — China is investigating DeepSeek and Moonshot, The Information reports - TNW ,The heart of tech™
- www.interconnects.ai — Debating RSI, the US-China Gap, and Jaggedness with JS Denain of Epoch AI
-
Anthropic Details Data Extraction Tactics Used by Chinese AI Firms
Chinese regulatory authorities are investigating DeepSeek and Moonshot AI following allegations that the laboratories secretly rerouted millions of user prompts through Anthropic's Claude systems to train domestic models. Anthropic published a report detailing 190 million extractions, revealing a specific translation prompt method that forced Claude to expose its internal deliberation processes. The Cyberspace Administration of China opened probes into DeepSeek and Moonshot AI after narrowing its initial inquiry from seven domestic companies to focus directly on these two labs over the alleged data harvesting.
Why it matters
This regulatory scrutiny highlights growing tensions around model distillation and intellectual property theft in the artificial intelligence sector. Chinese authorities stepped in after American competitors accused foreign labs of siphoning proprietary capabilities by passing third-party user data through Claude. The investigation tests how regulators handle cross-border data routing and model training practices.
What is confirmed
- China has opened investigations into DeepSeek and Moonshot AI after allegations that the firms secretly routed user requests to Anthropic's Claude models.
- Anthropic disclosed 190 million extractions involving a single line asking for a translation into katakana-only Japanese.
- The Cyberspace Administration of China shifted its focus to DeepSeek and Moonshot AI after an initial inquiry into seven firms.
What to watch next
- Official findings or penalties from the Cyberspace Administration of China regarding DeepSeek and Moonshot AI.
- Additional technical disclosures from Anthropic regarding Alibaba or other distillation campaigns.
confidence 100%Sources used for this update (2)
- note.com — 190 Million Extractions Disclosed by Anthropic: The Method Was a Single Line Asking for a Translation into Katakana-Only Japanese
- techstory.in — China Investigates DeepSeek, Moonshot AI Over Alleged Claude Data Rerouting
-
China investigates DeepSeek and Moonshot AI over Claude data routing
China's internet regulator is probing DeepSeek and Moonshot AI following accusations from Anthropic that the firms secretly routed millions of user exchanges through Claude to train their own models. The Cyberspace Administration of China shifted its focus to these two companies after an initial inquiry into seven firms. Authorities are investigating whether the labs siphoned capabilities from the American competitor by passing user prompts to Claude and returning those responses to their own users.
Why it matters
The probe focuses on distillation campaigns used to clone Claude models. While intellectual property theft is a common foreign concern, this investigation is driven by domestic Chinese authorities.
What is confirmed
- China's internet regulator is investigating DeepSeek and Moonshot AI.
- Anthropic accused DeepSeek and Moonshot AI of routing user exchanges through Claude to train their own models.
Still unconfirmed
- Moonshot AI's Kimi chatbot transferred questions from the Chinese military and Russian agencies to Claude.
What to watch next
- Findings from the Cyberspace Administration of China regarding data leaks
- Official responses from DeepSeek and Moonshot AI
confidence 90%Sources used for this update (2)
- decrypt.co — China Probes DeepSeek and Moonshot Over Alleged Data Leaks to Anthropic's Claude
- hashout.jp — Kimi・DeepSeekが中国軍やロシア機関の質問もClaudeに転送していた、中国当局が機密漏洩で動いた理由
-
China Probes DeepSeek and Moonshot Over Anthropic Claims
China is investigating domestic artificial intelligence developers DeepSeek and Moonshot AI following allegations that the companies siphoned capabilities from American competitor Anthropic. The Cyberspace Administration of China initiated the inquiry after Anthropic accused Chinese firms of executing massive distillation campaigns to clone Claude models. Regulators initially summoned representatives from seven named companies before narrowing the probe onto DeepSeek and Moonshot. The investigation focuses on whether these labs routed user prompts and data to Anthropic. While foreign accusations previously centered on intellectual property theft, the current pressure originates directly from domestic authorities.
Why it matters
The regulatory scrutiny arrives immediately ahead of planned artificial intelligence discussions between leaders Trump and Xi. Domestic oversight highlights Beijing's sensitivity regarding the methods utilized by local labs to achieve advanced reasoning capabilities. The probe marks a shift from international accusations to direct internal interrogation of top industry executives.
What is confirmed
- China's Cyberspace Administration is probing DeepSeek and Moonshot AI over accusations that the companies routed user prompts and data to Anthropic's Claude.
- Chinese regulators summoned representatives from all seven Chinese companies named in Anthropic's report before narrowing their focus to DeepSeek and Moonshot.
What to watch next
- Official findings or penalties from the Cyberspace Administration of China against DeepSeek or Moonshot AI.
- Developments or policy shifts emerging from the upcoming Trump-Xi AI talks.
confidence 100%Sources used for this update (2)
- finance.yahoo.com — DeepSeek, Moonshot Reportedly Face China Scrutiny Over Anthropic Claims Ahead Of Trump-Xi AI Talks
- gizmodo.com — China Probes DeepSeek, Moonshot AI Over Anthropic’s Claims They Route Requests to Claude
-
Anthropic reports 200 million distillation attempts by Chinese AI labs
Anthropic released a September 2026 report detailing massive efforts by Chinese AI firms to extract reasoning capabilities from Claude models. The company attributes over 151 million response exchanges to Qwen, over 23 million to Kimi, and over 12.1 million to DeepSeek within a 14-day window. These distillation campaigns aim to clone frontier model capabilities for use in domestic Chinese models. While Anthropic identifies these activities as unauthorized, third-party verification of these attributions remains unavailable and the Chinese government denies malicious intent.
Why it matters
Model distillation allows developers to train smaller models using the outputs of more powerful ones. This activity occurs amid diplomatic tension as Donald Trump and Xi Jinping prepare for meetings. The scale of these exchanges suggests a systemic effort to accelerate Chinese AI performance using US-developed intelligence.
What is confirmed
- Anthropic reports China-based AI companies attempted to extract Claude reasoning and capabilities.
- The total number of distillation exchanges from China-based labs reached nearly 200 million.
Still unconfirmed
- DeepSeek conducted over 12.1 million response exchanges in 14 days.
What to watch next
- Official responses from Alibaba, Moonshot AI, and DeepSeek regarding the specific exchange numbers.
confidence 80%Sources used for this update (2)
- digitalmarketreports.com — Anthropic Alleges China-Based AI Labs Conducted Nearly 200 Million Distillation Exchanges
- note.com — Claudeの「不正蒸留」報告は、APIの外側で何を問うのか
-
Anthropic Accuses Chinese AI Firms of 190 Million Claude Distillation Attacks
Anthropic reports that Chinese AI labs launched nearly 190 million distillation attacks against Claude models between May and July 2026. The company identifies Alibaba, Moonshot AI, DeepSeek, Zhipu, and Xiaomi as participants in a campaign to extract knowledge from its frontier models. Anthropic claims Moonshot and DeepSeek secretly routed real user requests through intermediate platforms to Claude to clone its capabilities. China has rejected these claims of malicious activity, though the government remains willing to hold AI talks with the US ahead of planned meetings between Donald Trump and Xi Jinping.
Why it matters
AI distillation involves using a powerful model to generate training data for a smaller, cheaper model. These accusations suggest a systemic effort to bypass US export controls by cloning frontier AI capabilities. The dispute adds tension to US-China diplomatic relations regarding technology security.
What is confirmed
- Anthropic accused Alibaba, Moonshot AI, DeepSeek, Zhipu, and Xiaomi of launching nearly 190 million distillation attacks against Claude between May and July 2026.
- Anthropic claims Moonshot and DeepSeek secretly routed user requests to Claude.
- China has denied US claims that its AI distillation efforts are malicious.
- China expressed willingness to hold AI talks with the US.
Still unconfirmed
- Distillation attacks on Claude grew 12-fold between May and July 2026.
- Six Chinese AI firms are aggressively copying US frontier models.
- Alibaba stock dropped 2.6% because the US views AI distillation as a security threat.
What to watch next
- Results of the planned Trump-Xi talks regarding AI security
- Official responses from Alibaba, DeepSeek, and Moonshot AI
- Further evidence of intermediate platforms used to route user requests
confidence 90%Sources used for this update (16)
- AP News — China hits back at US claims of ‘malicious’ AI distillation ahead of planned Trump-Xi talks
- Bloomberg.com — China ‘Willing’ to Hold AI Talks With US Despite Row Over Models
- arstechnica.com — Six Chinese AI firms accused of aggressively copying US frontier models
- TechCrunch — Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek
- Bloomberg.com — Moonshot Secretly Routed User Requests Through Claude, Anthropic Says
- WSJ — How Chinese AI Firms Tried to Clone U.S. AI Models
- CNBC — Chinese AI labs secretly used millions of Claude exchanges to train their models, Anthropic says
- Yahoo Finance — Alibaba Drops 2.6% as U.S. Turns AI Distillation Into a Security Threat
- South China Morning Post — Anthropic claims Moonshot, DeepSeek secretly diverted user requests to Claude
- Foundation for Defense of Democracies — How China Is Stealing America’s AI
- WSJ — Chinese AI Firms Try to Clone U.S. Rivals
- WSJ — The Morning Risk Report: How Chinese AI Firms Tried to Clone U.S. AI Models
Community Sentiment: How do you assess this situation?
Voice your perspective · Real-time aggregated sentiment from the Live Feeds community