AI penetration testing is a security assessment of the systems you build around artificial intelligence, particularly large language models. It focuses on the risks unique to AI, such as prompt injection, data leakage, and models being manipulated into unsafe behaviour, alongside the security of the application and infrastructure around them.