Mark Russinovich said at Build 2024 that AI is vulnerable to attacks such as prompt injections, data poisoning, and model misclassification.
