Tagged:

Fine-Tuning Trojans: Injecting Backdoors Through the Model Training Pipeline

How malicious training data, tampered datasets, and compromised fine-tuning APIs plant backdoored behavior in legitimate base models — and what defenders can do.