Over the past years, automated vulnerability prediction research, mainly supported by AI techniques, has grown in popularity. While a large number of studies have been proposed, they often make simplification assumptions, which limit their applicability and adoption. This talk will provide a historical view of the vulnerability prediction approaches and will focus on the challenges and limitation that they face through the lens of three different research communities, i.e., AI, SE and Security. The talk will conclude with a discussion on the links between vulnerability prediction and testing, showing potential applications and cross-fertilization between the two research fields.