Anthropic: Industrial-Scale Distillation Attacks
Source: https://x.com (tweet) Author: Anthropic (inferred) Date: 2025-2026
Summary
Anthropic identified industrial-scale distillation attacks on Claude from DeepSeek, Moonshot AI, and MiniMax. These labs created 24,000+ fraudulent accounts and generated 16M+ exchanges to extract Claude’s capabilities for training their own models.
Key Claims
- DeepSeek, Moonshot AI, MiniMax identified as attackers
- Scale: 24,000+ fraudulent accounts, 16M+ synthetic exchanges
- Goal: extract Claude capabilities for training competitor models
- Anthropic publicly identified the attacks
Entities
Concepts
- Synthetic Data — distillation as systematic synthetic data extraction
- AI Alignment — model IP and capability leakage