Backdoor Attack

Maliciously training models to behave normally except when specific triggers are present.

This concept is essential for understanding specialized ai topics and forms a key part of modern AI systems.