The Internet of Things (IoT) is being widely used in many application areas, including Smart Cities, Health Care, Agriculture, Power and Energy, Industries, etc. Due to the wide applications of IoT networks, the number of vulnerabilities is also increasing significantly. Securing an IoT network from different attacks and vulnerabilities is an important research problem. As the Deep Learning (DL) methods gain popularity in handling massive amounts of data, the problem of IoT network attack identification can be handled using DL methods. In this paper, we introduced a Multi-task Learning (MTL) model that performs two tasks parallelly, viz., (i) binary classification and (ii) multiclass classification by sharing some informative features during model training. The proposed MTL model uses an Autoencoder (AE), where an embedded latent layer works as a feature extractor to reduce the dimensionality of the input data. The beauty of the proposed model is the use of a shared dense layer that selects the informative features responsible for binary and multiclass classification. The model is applied on three benchmark datasets and observed high classification accuracy (ACC) as 99.99%, 99.98%, and 99.89% for binary classification on UNSW-BoT-IoT, N-BaIoT, and Edge-IIoT, respectively.