This question drives the work of the Applications, Platforms and Systems Security (APSS) research group. AI is now part of the mobile devices, cyber-physical systems and generative models that billions of people rely on, and its failures can cause privacy, security and safety harms at that scale. We study these systems as an adversary would, combining large-scale empirical measurement, adversarial modelling and diffusion-based generative analysis to find and characterise new attack surfaces. We then design safeguards whose decisions people can verify, understand and contest, using explainability, concept unlearning and neurosymbolic methods such as computational argumentation. We collaborate across disciplines and with industry and government, so that these methods reach the computing environments people use every day.