From the course: Complete Guide to Cybersecurity: A Practical Approach

Unlock this course with a free trial

Join today to access over 25,500 courses taught by industry experts.

Explaining the model theft attack

Explaining the model theft attack

From the course: Complete Guide to Cybersecurity: A Practical Approach

Start my 1-month free trial Buy for my team

Explaining the model theft attack

“

- Let's go over another type of AI attack called the model theft attack. This attack is also often referred to as model extraction. And basically, this is when the attacker tries to create a copy of the target's model without directly accessing it or directly accessing the parameters or the training data within that model. And then basically, this is done using the target's prediction API to make queries, and then the responses to these queries are used to train a similar model that the attacker, of course, controls. Now, the attacker has black box access, so basically doesn't know a lot about the underlying infrastructure of the AI system. So they can provide inputs to the model and observe the outputs coming back to the attacker. But if they cannot directly observe the model parameters or the internal workings, then it can infer or create their own model to actually, of course, perform the different attacks. So these are the steps. The first thing is that you query the target model,…

Contents