AI systems are revolutionising industries, driving advancements in areas like smart manufacturing and autonomous vehicles, but it's important to understand that they are not a panacea. While AI technologies have shown promise in detecting and preventing cyber-attacks, they also have various limitations. Adversarial attacks can exploit vulnerabilities within AI models and corrupt the data that AI models learn from, causing these models to generate inaccurate outputs. This talk will shed some light on the key security challenges associated with the widespread use of AI systems, and the appropriate countermeasures that can be applied guided by past research experience in the field.