Adversarial Attack

The deliberate manipulation of inputs to an AI system to cause it to make errors, misclassify or behave unexpectedly, often through imperceptible perturbations.

Deliberately tricking an AI by giving it sneaky inputs. A classic example: putting a tiny sticker on a stop sign that's invisible to humans but causes a self-driving car's AI to misread it.

Adversarial attacks represent a real and growing threat to AI systems. Your AI risk management framework must include threat assessments for adversarial attacks, particularly for AI systems that make high-stakes decisions.