The National Cyber Security Centre (NCSC) has published a paper on adversarial machine learning (AML) attacks that outlines attack classes exploiting ML-specific vulnerabilities, model behaviours and information leakage across the model lifecycle, including development, training and deployment, and highlights risks associated with large model sizes, open-source components and an expanded attack surface. The paper groups AML techniques into categories such as model characterisation, model inversion, training data poisoning, malicious model training, model input manipulation, model artefact manipulation and model hardware attacks, and explains how these techniques may enable malicious actors to achieve objectives including reconnaissance, performance degradation, resource exhaustion, output attribution, embedding hidden behaviours, evading detection, data extraction and unauthorised access. It distinguishes AML attacks from traditional cybersecurity threats while noting that ML systems remain susceptible to both and emphasises the need for robust security aims to protect confidentiality, integrity and performance. The paper aims to raise awareness, support threat modelling, establish a common language for ML security and highlight research gaps to improve collaboration and the development of defences against evolving AML attacks.