Training parameterised quantum circuits (PQCs) is hindered by the high measurement cost of gradient estimation, which increases linearly with the number of trainable parameters. To address this, researchers have developed a framework for forward gradient estimators, leveraging the forward mode of automatic differentiation. This approach enables more efficient estimation of gradients in PQCs, reducing the shot budget required for training. The proposed framework has significant implications for the development of quantum machine learning models, particularly in the context of state-aligned activity involving PQCs, which shifts the threat model from criminal to geopolitical1. This change in threat model necessitates a different approach to security, one that takes into account the unique characteristics of quantum computing and the potential for nation-state actors to exploit vulnerabilities. As a result, practitioners must reevaluate their security protocols to address the emerging risks associated with PQCs.