AI systems can be attacked, and these attacks can expose sensitive data or compromise the model itself.

This demo shows the main privacy risks in building AI systems, especially when models are trained, shared, or queried.

Privacy-preserving techniques (such as noise-based protection, secure collaboration, and encrypted computation) can reduce these risks.

The demo also highlights the practical trade-offs AI operators need to consider, including accuracy, efficiency, robustness, fairness, and explainability.