As one of data sources in cloud–edge–terminal collaboration enabled artificial intelligence of things (CETC-AIoT), the integrity and confidentiality of sensed information in the terminal side directly impact on the modeling and decision-making for CETC-AIoT. However, due to openness of transmission media among cloud, edge and terminal, it could be vulnerable to pollution attack and eavesdropping attack. Additionally, the constrained resources of some terminals make it difficult to deploy strong security schemes. In this context, how to make a tradeoff between the requirement of security and the limitation of resources needs to be explored. Therefore, toward secure and lightweight data transmission for CETC-AIoT, we propose a novel Gold sequence-based secure network coding (GS-SNC) scheme in this article. Specifically, the Gold sequence is introduced to generate the pseudo-random sequence, which is used to scramble and descramble the original information. The precoding matrix is constructed to encode and decode the scrambled information. The intermediate nodes perform the random linear network coding. The simulation results show that GS-SNC has advantages compared with double prime numbers-based secure network coding (DP-SNC) and secure practical network coding (SPOC), in terms of security, computational complexity, encryption capacity, and space overhead.