Anthropic Says Rival AI Labs Used Claude as a Training Ground
Synopsis
Anthropic reported nearly 200 million Claude exchanges linked to alleged AI distillation campaigns, while US agencies named six Chinese companies and Stanford reported a narrowing US-China model performance gap.
Anthropic claims that its rival AI labs have been using Claude heavily to steal its capabilities and create training data for their models in a game of cat and mouse on state-of-the-art AI systems.
Anthropic said in its latest Threat Intelligence report that it had identified nearly 200 million exchanges associated with five separate distillation campaigns. The activity included an effort to replicate skills such as coding, data analysis, logical reasoning and the utilization of AI tools.
While there have been myriad ways to develop AI, one that is legit is distillation, which entails training another system using the output of another more capable system.
Anthropic said that the observed behavior was not only that, with users reportedly using a huge number of accounts and carefully crafted prompts to try to extract capabilities from Claire.
Alibaba Campaign Accounted for 151 Million Exchanges
Anthropic matched the biggest campaign to Alibaba, that they claimed had over 151 million exchanges with Claude from May-July 2026.
Anthropic labeled the activity as fraudulent, with nearly 3 million exchanges daily from over 3,500 accounts. The effort aimed to gather the necessary training data to refine the Alibaba's Qwen models, the company said.
Anthropic also accused Moonshot AI and DeepSeek of redirecting customer interactions to Claude and extracting training data from the outputs. Anthropic said that a separate campaign by Moonshot made almost 300,000 requests over 10 days from almost 5,000 accounts.
The claims follow up on an earlier report last month by Anthropic of over 16 million Claude exchanges made by DeepSeek, Moonshot and MiniMax via approximately 24,000 accounts.
US Agencies Broaden the Distillation Dispute
The dispute has also seeped into the US government. CISA, NSA and FBI issued a joint cybersecurity advisory on Thursday identifying six China-based companies, including DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.Through extensive distillation projects, AI, and said billions of tokens had been harvested from the top-tier models available in the United States.
The agencies said that the activity had been running since at least late 2024 and was likely to have been done with the knowledge of the Chinese government. The Chinese government have denied the allegations saying that they are baseless and that distillation is a commonly used technological process.
The dispute is occurring because the United States continues to have a substantial investment edge in AI. The Stanley's 2026 AI Index estimated the US private AI investment at $285.9 billion in 2025 but said that China's government-backed funding also wasn't included in the total of the private funding figures.
Concurrently, Stanley discovered that the difference in the performance of the top U.S. and Chinese AI models was 2.7% by March 2026, underscoring the importance of the access to sophisticated AI models in the global AI race.
Source: TechCrunch
Pooja Malik is a business journalist with over six years of experience covering startups, entrepreneurship, and emerging trends. She has previously worked with leading media platforms such as YourStory Media and BW BusinessWorld, where she reported on business, policy, and market developments. Currently, she serves as Editor at The Inspirepreneur Magazine, where she writes and edits stories across business, lifestyle, and travel, with a focus on clarity, accuracy, and reader relevance.
You Might Also Like
Netflix names longtime Director Jay Hoag Chairman following Hastings exit
Mark Allison: The Man Who Turned a Failing Elders Group Into a $2 Billion Business
