The record
Written from the 1 report below. Nothing here is unsourced.
- Anthropic says it identified and disrupted industrial-scale illicit distillation attacks against Claude from seven China-based AI labs, including Alibaba, Moonshot, DeepSeek, Z.ai, and MiniMax.
- The labs allegedly used fake accounts, stolen credit cards, and proxy networks to extract Claude's capabilities, with the largest campaign involving 151 million exchanges attributed to Alibaba-affiliated operators.
- Some labs, including DeepSeek, Xiaomi, and Moonshot, rerouted their own users' requests to Claude without permission and used the responses as training data, including exchanges containing sensitive information.
- Anthropic says it is fighting back by banning reseller accounts, summarizing Claude's internal reasoning before responding, and encrypting reasoning to make stolen transcripts less useful.
- The episode matters because it shows frontier AI models are being targeted at scale for unauthorized copying, raising questions about protecting user conversations and AI intellectual property.
What to watch next
- Whether other frontier labs like OpenAI and Google report similar distillation campaigns and coordinate defenses.
- Any response or denial from the named Chinese companies, including Alibaba, DeepSeek, and Moonshot.
- Whether Anthropic's technical fixes, such as encrypted reasoning and preserved thinking, actually reduce future distillation attacks.
Coverage1
1 report
English national1
All filed from India
Named China · United States · Singapore · Japan · Alibaba · Anthropic · DeepSeek · Google · MiniMax · Moonshot AI · OpenAI · SenseTime · Xiaomi · Zhipu
The 1 report is listed beside the record.
Ask this story
Answers cite the reports above, or say they can't.