Low-latency and high-reliability communication is one of the crucial building blocks for many Internet-of-things (IoT) applications. It is observed that many new applications raise a sporadic traffic with short messages, in which case considerable amount of communication resource is dominated by control data. A joint blind deconvolution and blind demixing approach has recently been proposed to reduce the signaling overhead by allowing the recovery of multiple pairs of signal and channel based on only one received signal. However, the existing approach is not capable to verify its own recovery performance, making it vulnerable to jamming attacks. In this paper, we introduce a novel scheme for stable signal recovery under jamming attacks. Our method of adding a guarding node to the network is able to detect the jamming attack, reject the jamming signals and recover the legitimate signals. We also propose an iterative reweighted optimization algorithm that supports automatically learning of the optimal regularization parameter. Numerical experiments demonstrate that the proposed approach achieves about 4 dB reduction in average recovery error.