Privacy-enhancing technologies are crucial in mitigating privacy risks associated with machine learning systems. Technologies like secure multiparty computation, homomorphic encryption, secure enclaves (also known as trusted execution environments or confidential computing) and differential privacy provide ways to secure data while maintaining its utility. These technologies enable the safe application of machine learning by protecting sensitive information, even as it is processed and analysed. Typically falling into either input privacy (protecting the inputs to a function) or output privacy (preventing the reversibility of inputs from the outputs) categories, they serve as a bridge between the technical requirements of data analysis and the legal imperatives of privacy protection, ensuring that machine learning systems can operate within the bounds of privacy regulations.