To implement ubiquitous intelligence of the sixth generation mobile networks (6G), space-air-ground integrated networks (SAGIN) have emerged as a promising infrastructure to provide globally seamless coverage and powerful cloud computing services. With SAGIN, Internet of Things (IoT) devices can offload compute-intensive tasks to unmanned aerial vehicles (UAVs) and satellites. However, due to the vulnerability of wireless communications to eavesdropping attacks, privacy-preserving task scheduling in SAGIN is of great challenge. To address this problem, we propose a privacy-driven multi-agent proximal policy optimization-based security-aware SAGIN task scheduling mechanism, where we jointly optimize the delay, energy consumption and security utility of tasks considering different privacy protection needs. In this mechanism, privacy protection is accomplished from two aspects. First, the privacy level of tasks and the privacy leakage risks of the communication link are fully considered in the selection of the scheduling destination. Second, we employ cryptography techniques and conceive an optimal encryption configuration scheme to achieve the most cost-effective privacy protection. Moreover, a convex optimization method is utilized to allocate CPU-cycle frequencies for computing services. In the end, simulation results demonstrate that our proposed algorithm outperforms other benchmarks in terms of convergence speed, offloading overhead and security utility of IoT devices.