Anthropic Alleges Massive Claude Copying Operation by Chinese AI Labs

Anthropic claims several Chinese AI companies launched large-scale efforts to replicate its Claude model capabilities. The allegations come as the company faces increasing geopolitical scrutiny.

Anthropic says three Chinese AI labs, MiniMax, DeepSeek, and Moonshot, ran coordinated campaigns designed to extract Claude’s advanced capabilities at scale.

Alleged Large-Scale Account Network Targeted Claude Capabilities

Anthropic describes the operation as “industrial scale” distillation. The company alleges that around 24,000 fraudulent accounts generated more than 16 million interactions with Claude.

The campaigns reportedly focused on advanced capabilities such as agentic reasoning, tool use, and coding performance.

Anthropic claims the activity relied on “hydra cluster” account networks that rotated infrastructure to evade detection systems.

Breakdown of Alleged Activity

Anthropic provided a detailed breakdown of the suspected activity:

• MiniMax reportedly generated more than 13 million exchanges targeting agentic coding and tool use capabilities.
• Moonshot allegedly conducted around 3.4 million exchanges focused on reasoning, coding, and computer vision.
• DeepSeek is said to have produced roughly 150,000 exchanges aimed at reasoning, grading systems, and censorship-safe responses.

Distillation: Legitimate Practice or Competitive Shortcut?

Model distillation is a common AI technique used internally to create smaller and more efficient models from larger systems. Anthropic argues the issue arises when external competitors use mass query extraction to replicate proprietary capabilities at lower cost.

The company says it identified the actors with high confidence using IP correlation, metadata tracking, infrastructure signals, and corroboration from partners.

New Countermeasures Deployed

In response, Anthropic says it has deployed new classifiers and behavioral fingerprinting systems to detect distillation-style attacks.

The company is also sharing technical indicators with other AI labs and cloud providers to limit similar activity.

Additional safeguards include stronger verification processes for education and research accounts, along with improved detection for coordinated multi-account networks.

Broader Pressure on Anthropic

The claims surfaced shortly after reports that Anthropic faced scrutiny related to supply chain concerns in U.S. defense circles.

At the same time, the company continues to push forward with product updates, recently releasing Claude Sonnet 4.6 and Claude Opus 4.6 with performance improvements across reasoning and coding tasks.

The situation highlights growing tension in the global AI race, where model capability, infrastructure access, and intellectual property protection are becoming increasingly sensitive issues.

Via Neowin