INCIDENT MAP

Browse / Model Theft and Extraction / AML.CS0056

Model Distillation Campaigns Targeting Anthropic Claude

AML.CS0056 · 2026-02-23 case study

Class
Model Theft and Extraction
Technique
Exfiltration via AI Inference API
Also involves
Direct Prompt Injection, Misinformation and External Harm, AI Framework and Platform Flaws
Target
Anthropic Claude (Incident)

How it worked

  1. Resource Development · Acquire Infrastructure: AI Service Proxies

    DeepSeek, Moonshot AI, and MiniMax used commercial proxy services to gain access to Claude. This circumvented Anthropic's policy of not offering commercial access to Claude in China.

  2. Initial Access · Valid Accounts

    DeepSeek, Moonshot AI, and MiniMax accessed Claude through approximately 24,000 fraudulent accounts collectively, including accounts supplied through commercial proxy networks. These accounts enabled authenticated requests to Claude through Anthropic API and third-party cloud platforms.

  3. AI Attack Adaptation · LLM Prompt Crafting

    DeepSeek, Moonshot AI, and MiniMax crafted large datasets of prompts designed to elicit responses with content that is useful in distilling capabilities from Claude.

  4. Defense Evasion · Centralized AI Service Request Routing

    DeepSeek, Moonshot AI, and MiniMax used the commercial AI service proxies to distribute coordinated requests across accounts and access pathways to evade detection.

  5. Exfiltration · Exfiltration via AI Inference API

    DeepSeek, Moonshot AI, and MiniMax obtained Claude-generated responses through API access, including access mediated by commercial proxy services. The responses were transferred to the labs for use in their distillation campaigns.

  6. Collection · Outputs from AI Models

    DeepSeek, Moonshot AI, and MiniMax gathered Claude's responses as supervision for training their own models. Across the campaigns, Anthropic reported more than 16 million exchanges.

  7. Resource Development · Replicate AI Model Behavior: Replicate Behavior from Model Outputs

    DeepSeek, Moonshot AI, and MiniMax used responses generated by Claude to train their own models and acquire selected Claude capabilities through distillation.

  8. Impact · External Harms: AI Intellectual Property Theft

    DeepSeek, Moonshot AI, and MiniMax acquired Claude capabilities via distillation at a fraction of the cost of developing their own models. They targeted the source model most differentiated capabilities, including agentic reasoning, tool use, and code generation.

  9. Impact · External Harms: Societal Harm

    Anthropic warned that models developed through illicit distillation could lack safeguards that restrict dangerous capabilities. Such models could support offensive cyber operations, disinformation campaigns, mass surveillance, or other activities that harm the public or vulnerable groups.

  10. Impact · External Harms: User Harm

    Anthropic warned that models developed through illicit distillation were unlikely to retain the safeguards of the source model. Deployment of these models could expose individual users to harmful outputs or enable misuse against individuals.

Summary

Anthropic uncovered campaigns to extract Claude's capabilities carried out by the three Chinese AI Labs: DeepSeek, Moonshot, and MiniMax. Collectively, these campaigns used approximately 24,000 accounts and 16 million queries. They used model distillation to train their own models on the outputs of Claude in an attempt to replicate Claude's capabilities such as agentic reasoning, code generation, tool use, and computer use.

The labs obtained access through commercial proxy services that circumvented Anthropic policy on commercial access to Claude in China. Anthropic described proxy networks that distributed requests across accounts, its API, and third-party cloud platforms. These networks replaced banned accounts and mixed distillation requests with unrelated customer traffic, complicating detection of coordinated activity.

The following table summarizes the capabilities each lab targeted and the scale of their campaigns:
| Lab | Capabilities targeted for distillation | Reported scale |
|---|---|---|
| DeepSeek | Reasoning across multiple tasks, rubric-based evaluation, censorship-safe alternatives to politically sensitive queries | Over 150,000 exchanges |
| Moonshot AI | Reasoning and tool use, coding and data analysis, computer-use agent capabilities, computer vision | Over 3.4 million exchanges |
| MiniMax | Agentic coding, tool use, orchestration | Over 13 million exchanges |

Anthropic characterized these campaigns as illicit capability extraction that reduced the time and cost required to develop competing models. It also warned that models developed through illicit distillation could lack safeguards and enable downstream misuse, including offensive cyber operations, disinformation, and mass surveillance.

References

Source: MITRE ATLAS case studies, used under Apache 2.0.