BREAKING
AI Rivals Unite vs Model-Copying
How Adversarial Distillation Works
1
Fake accounts
↓
2
Mass API queries
↓
3
Extract abilities
↓
4
Train copy
0
Fake accounts
0
M
Interactions
0
Accounts per network
Exchanges Tied to Chinese Firms
MiniMax
13
Moonshot AI
3.4
DeepSeek
0.15
Legitimate vs Adversarial
Distillation
Legal
●
Teacher trains student
●
Widely used
●
Efficient
Adversarial
Alleged abuse
●
Fake accounts
●
Bypass limits
●
Security risk
Boundary Now Up to Regulators
AI NEWS BLITZ
OpenAI, Anthropic and Google are teaming up to fight alleged Chinese model-copying.