Adversarial Robustness

The ability of an AI model to maintain correct performance when subjected to adversarial attacks or intentionally crafted malicious inputs.

How well an AI holds up against tricks and attacks. A robust model won't be easily fooled by small changes to inputs that wouldn't confuse a human.

Robustness against adversarial attacks should be a requirement in your AI governance standards. AI systems deployed in critical applications must be tested for adversarial resilience before and after deployment.