BREAKING
LLMs Used to Unmask Anonymous Posters
0
%
recall at 90% precision
0
%
prior methods
0
$
cost per target
The Four-Stage Attack Pipeline
1
Extract
↓
2
Search
↓
3
Reason
↓
4
Calibrate
Why This Differs From Old Attacks
Classic Re-ID
e.g. Netflix
●
Structured data
●
Manual effort
LLM Method
New
●
Unstructured forum text
●
Minutes to hours
Practical Obscurity Erodes
Rethink the Threat Model
AI NEWS BLITZ
Researchers from ETH Zurich and Anthropic show how AI can identify anonymous online users.