llms.blog
AI

Anthropic accuses Alibaba of largest Claude distillation campaign to date

# Anthropic accuses Alibaba of running largest Claude distillation campaign to date Anthropic has accused Alibaba of orchestrating the largest model distillation campaign it has ever publicly disclosed, according to a letter the company sent to the US Senate Banking Committee on June 10, 2026. The letter, first reported by Reuters on June 24, alleges that operators affiliated with Alibaba and its Qwen AI lab generated more than 28.8 million exchanges with Claude through nearly 25,000 fraudulen

axc
axc
3 min
Anthropic accuses Alibaba of largest Claude distillation campaign to date

# Anthropic accuses Alibaba of running largest Claude distillation campaign to date

Anthropic has accused Alibaba of orchestrating the largest model distillation campaign it has ever publicly disclosed, according to a letter the company sent to the US Senate Banking Committee on June 10, 2026.

The letter, first reported by Reuters on June 24, alleges that operators affiliated with Alibaba and its Qwen AI lab generated more than 28.8 million exchanges with Claude through nearly 25,000 fraudulent accounts between April 22 and June 5, 2026. The exchanges targeted Claude's software engineering, agentic reasoning, and cybersecurity capabilities, the same abilities embodied in Anthropic's Mythos Preview model.

Distillation is a well-known machine learning technique where a smaller, less capable model is trained on the outputs of a larger, more capable one. It has legitimate uses: companies routinely distill their own large models into smaller, faster variants. The controversy arises when the technique is used to copy a competitor's model without authorization, effectively extracting intellectual property through API access.

According to Anthropic, the Alibaba operation was designed to mimic normal user behavior at industrial scale. The 25,000 accounts were spread across geographically dispersed servers to avoid triggering rate limits or detection systems. At 28.8 million exchanges over roughly six weeks, the campaign would have harvested an estimated 57.6 billion tokens of Claude output.

Anthropic told senators that distillation accelerates China's ability to reach the capabilities of its most advanced models at a fraction of the development cost. The company called on Congress to implement information-sharing frameworks, penalties for foreign entities engaged in industrial-scale siphoning, and cooperation from cloud providers to detect and block suspicious traffic patterns.

Alibaba has denied the allegations. The company did not respond to requests for comment from Reuters or the Wall Street Journal at the time of publication. Alibaba's AI business has been growing rapidly; CEO Eddie Wu said on the company's fiscal Q4 2026 earnings call that cloud unit external revenue grew 40% year-over-year, with AI-related products accounting for 30% of that total.

The accusation mirrors a similar claim by OpenAI, which told House lawmakers earlier this year that DeepSeek used distillation to improve its models on OpenAI outputs. Google's Threat Intelligence Group has also warned of rising distillation attacks targeting Western AI companies.

The Forbes analysis published August 3 frames the incident as part of a broader pattern: Chinese AI firms accused of copying American models, investors pushing valuations sky-high regardless, and US policymakers struggling to define what constitutes IP theft when the stolen asset is a statistical model's output rather than source code.

Anthropic's letter was addressed to Senator Tim Scott (R-SC), chair of the Banking Committee, and Senator Elizabeth Warren (D-MA), its ranking member. No legislation has been introduced in direct response to the disclosure.

## Sources

- [Reuters: Anthropic says Alibaba illicitly extracted Claude AI model capabilities](https://www.reuters.com/world/china/anthropic-says-alibaba-illicitly-extracted-claude-ai-model-capabilities-2026-06-24) - [WSJ: Anthropic Accuses Chinese Companies of Siphoning Data From Claude](https://www.wsj.com/tech/ai/anthropic-accuses-chinese-companies-of-siphoning-data-from-claude-63a13afc) - [Forbes: Chinese AI Firm Siphoned American AI Knowledge From Anthropic Claude By Using Millions Of Prompts](https://www.forbes.com/sites/lanceeliot/2026/08/03/chinese-ai-firm-siphoned-american-ai-knowledge-from-claude-by-using-millions-of-prompts-and-trained-on-the-responses) - [Barchart: Anthropic Accuses Alibaba of Massive AI Data Distillation Attack](https://www.barchart.com/story/news/3038274/anthropic-accuses-alibaba-of-massive-ai-data-distillation-attack-how-to-play-baba-stock-here) - [Digital Applied: Anthropic Accuses Alibaba of a Record Distillation Campaign](https://www.digitalapplied.com/blog/anthropic-alibaba-distillation-campaign-2026-ai-ip-war)

AILLM
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Disaggregated Prefill and Decode in Production LLM Serving: Architecture, Network KV Cache Migration, Chunked Prefill Trade-Offs, and Asymmetric Hardware Economics

    Large language model serving systems have historically treated transformer execution as a homogeneous sequence of forward passes over a single unified GPU pool. Under continuous batching engines, incoming requests execute their prompt evaluation (prefill) and autoregressive token generation (decode) on the exact same accelerators, co-locating both phases within shared iteration batches. While continuous batching improves GPU compute utilization compared to static batching, co-locating prefill a

    1 min

  • Direct Preference Optimization: Mathematical Derivation, Implicit Reward Formulation, and the Mechanics of RL-Free Alignment

    Direct Preference Optimization: Mathematical Derivation, Implicit Reward Formulation, and the Mechanics of RL-Free Alignment Aligning autoregressive large language models with human preferences has traditionally relied on Reinforcement Learning from Human Feedback (RLHF). In the standard formulation popularized by InstructGPT and related post-training regimes, alignment requires a multi-stage pipeline: supervised fine-tuning (SFT), training a separate reward model on pairwise comparison data, a

    1 min

  • Hugging Face Introduces gr.Workflow to Turn AI Pipelines into Visual Graphs and REST APIs

    Hugging Face has released gr.Workflow, a native extension to the Gradio framework designed to convert multi-stage artificial intelligence pipelines into interactive node graphs, visual user interfaces, and deployable REST APIs. Modern machine learning applications increasingly rely on compound pipelines that chain heterogeneous models: generating text via large language models, feeding prompts into diffusion systems, processing outputs through background removal or audio synthesis models, and a

    1 min