Machine learning (ML) and deep learning (DL) have spread across various fields, including medicine, finance, and networks, due to their capability of processing large data and learning valuable patterns. With a growth in usage, however, so have security threats. ML models have a predictable lifecycle, every stage of which introduces potential threats which can be exploited by an adversary in an attempt to breach confidentiality, integrity, or availability. The attacks can take forms of intercepting, manipulating, or inferring secret data, posing substantial security threats. This survey provides an overview of major security threats on ML models, including adversarial attacks, model extraction, membership inference, and backdoor attacks. We identify the implications of these threats and the urgent necessity for secure defense mechanisms for securing AI systems.