Prompt injection gets all the attention, but I think the bigger risks are in training data poisoning and model supply chain attacks. What security risks in AI systems do you think are underappreciated?
Sign in to answer this question.