GLOSSARY > MODEL INVERSION ATTACK
Model Inversion Attack
A cryptographic attack where adversaries reconstruct the private training data used to train a machine learning model by exploiting the model's confidence scores.
If an LLM is trained on sensitive enterprise IP, adversaries can systematically prompt the model to regurgitate the original training documents. Defending against Model Inversion requires strict Output Sanitization, Differential Privacy during training, and rigid secure enterprise boundaries to ensure the model cannot be exploited by unauthorized endpoints.
Explore More