# gradient-manipulation

Training Loop Vulnerabilities

Attacks on the training process itself including gradient manipulation, loss function tampering, learning rate schedule attacks, and training infrastructure compromise.