Adversarial Examples (d88c17bf-39b3-549b-af8d-ca1a97fed53f)
Adversarial examples are a type of evasion attack where malicious inputs are deliberately crafted to mislead AI models. These inputs are minimally modified, often imperceptible to humans, but can cause the model to produce incorrect or harmful predictions. Examples include researchers demonstrating that carefully designed patterns on accessories, like sunglasses, could deceive facial recognition systems into misidentifying individuals. Such examples are particularly problematic in critical domains like healthcare, finance, and security, where incorrect predictions could lead to severe consequences.
Threat-modeling question: Are we protected from adversarial examples?
| Cluster A | Galaxy A | Cluster B | Galaxy B | Level |
|---|---|---|---|---|
| Adversarial Examples (d88c17bf-39b3-549b-af8d-ca1a97fed53f) | PLOT4ai | Craft Adversarial Data (a7c30122-b393-4265-91b7-57cd1211e3f9) | MITRE ATLAS Attack Pattern | 1 |