Black-Box Transfer
Black-Box Transfer is a technique where a hacker builds a 'practice' version of an AI to figure out its weaknesses. Because the practice AI acts like the real one, the hacker can test their tricks on their own copy first, knowing that if the trick works on the practice model, it will likely work on the real one too.
In Black-Box Transfer attacks, the adversary uses one or more proxy models (trained via Create Proxy AI Model or Train Proxy via Replication) they have full access to, which serve as surrogates for the target model. By applying White-Box Optimization to these proxies, the adversary generates adversarial examples that exploit the shared decision boundaries between the proxy and the target, allowing the attack to generalize to the target system.
In Black-Box Transfer attacks, the adversary uses one or more proxy models (trained via Create Proxy AI Model or Train Proxy via Replication) they have full access to and are representative of the target model. The adversary uses White-Box Optimization on the proxy models to generate adversarial examples. If the set of proxy models are close enough to the target model, the adversarial example should generalize from one to another. This means that an attack that works for the proxy models will likely then work for the target model. If the adversary has AI Model Inference API Access, they may use Verify Attack to confirm the attack is working and incorporate that information into their training process.