131 characters, too long. Need 60-90. Let's shorten. "Anthropic sues DeepSeek, Moonshot
Photo by Mohamed Marey (unsplash.com/@abushihabmarey) on Unsplash
Anthropic says Chinese labs Deepseek, Moonshot and MiniMax launched a distillation attack on its Claude model, using over 24,000 fake accounts to submit more than 16 million queries, the‑Decoder reports.
Quick Summary
- •Anthropic says Chinese labs Deepseek, Moonshot and MiniMax launched a distillation attack on its Claude model, using over 24,000 fake accounts to submit more than 16 million queries, the‑Decoder reports.
- •Key company: Anthropic
- •Also mentioned: DeepSeek, MiniMax, IBM
Anthropic filed a lawsuit on Thursday alleging that three Chinese AI labs—DeepSeek, Moonshot and MiniMax—conducted an “industrial‑scale” distillation attack on its Claude model, according to a report by The Decoder. The complaint says the labs created more than 24,000 fraudulent accounts and submitted over 16 million queries to Claude, extracting reasoning steps, programming outputs and tool‑use behavior.
The Decoder details that DeepSeek focused on Claude’s chain‑of‑thought reasoning, harvesting “reward‑model data for reinforcement learning” and “censorship‑compliant answers on politically sensitive topics” across roughly 150,000 queries. Moonshot’s effort targeted agent‑based reasoning, tool usage, data analysis, computer‑vision tasks and even reconstruction of Claude’s thought processes, accounting for more than 3.4 million exchanges. MiniMax reportedly ran 13 million queries, emphasizing agent‑based programming and orchestration, and pivoted to a newer Claude version within 24 hours of its release.
Anthropic says the labs used proxy services to bypass China’s internet restrictions, allowing the fake accounts to reach Claude’s API despite the country’s firewall. The company alleges the attacks were designed to train smaller, domestically‑developed models by mimicking Claude’s performance on complex tasks, a practice known as model distillation.
Tom’s Hardware echoed Anthropic’s claim, describing the operation as “large‑scale distillation” that leveraged the 16 million exchanges to improve the Chinese developers’ own models. The outlet did not provide independent verification but cited Anthropic’s filing as the source.
TechCrunch noted that the accusations come amid heightened U.S. scrutiny of AI technology transfers, suggesting the lawsuit could intensify diplomatic pressure on China’s frontier‑AI sector. VentureBeat, meanwhile, highlighted that Moonshot and DeepSeek have recently launched competitive open‑source models, underscoring the strategic stakes of the alleged data theft.
Sources
- Hacker News Front Page
This article was created using AI technology and reviewed by the SectorHQ editorial team for accuracy and quality.