Export
The link always opens the latest version of this page.
Ask an AI about it
Open in ChatGPTA new chat that reads this entry from the site Open in ClaudeThe same, in ClaudeShare the link instead of a file, and they will always see the latest version.
Files are a snapshot of today. The live page has anything newer.
China-based AI model extraction campaign
A coordinated effort attributed to China to illicitly obtain proprietary capabilities and training data from advanced U.S. AI models.
Overview
The campaign emerged in October 2024 1 with the stated objective of extracting proprietary AI model capabilities and functionalities from U.S. frontier models 2. A secondary goal involved extracting billions of tokens from these models to reduce training costs and accelerate domestic AI development 1.
Objectives
Threat actors sought to compromise the intellectual property embedded in leading-edge American AI systems. The extraction of large token volumes represents a cost-reduction strategy, as generating equivalent training data through legitimate means would require substantial computational resources and time. By obtaining pre-trained model outputs, Chinese actors could potentially bootstrap their own model development while bypassing expensive inference and fine-tuning phases.
Attribution
The campaign is attributed to China with high confidence 2.
Facts 2
| Attribute | Value | Confidence | Sources | Cited |
|---|---|---|---|---|
| First seen | 2024-10 | low | 1 | 1 |
| Objective | Extract billions of tokens from frontier AI models to reduce training costs and accelerate AI development
also reported: Extract proprietary AI model capabilities and functionalities from U.S. frontier models (low)
|
low | 1 | 1 |