Adversarial machine learning

Techniques that fool, corrupt or extract information from machine learning models, and the defences against them.

Covers the field as a whole: attacks that fool, corrupt or extract information from machine learning models, including crafted adversarial inputs, and the defences against them. Corrupting the data a model learns from is Data poisoning; input that makes a language model follow an attacker's instructions is Prompt injection.

Scheme

• Technologies and capabilities

Broader terms

• AI security