Motivation: Electroencephalography (EEG) recorded during Transcranial Alternating Current Simulation (tACS) is highly desirable in order to investigate brain dynamics during stimulation, but is corrupted by large amplitude stimulation artefacts. Artefact removal algorithms have been presented previously, but with substantial debates on their performance, utility, and the presence of any residual artefacts. This paper investigates whether machine learning can be used to validate artefact removal algorithms. The postulation is that residual artefacts in the EEG after cleaning would be independent of the experiment performed, making it impossible to differentiate between different parts of an EEG+tACS experiment, or between different behavioural tasks performed. Methods: Ten participates undertook two tasks (nBack and backwards digital recall) during simultaneous EEG+tACS, exercising different aspects of working memory. Stimulation during no task and sham conditions were also performed. A previously reported tACS artefact removal algorithm from our group was used to clean the EEG and a Linear Discriminant Analysis was trained on the cleaned EEG to differentiate different parts of the experiment. Results: Baseline, baseline during tACS, working memory task without tACS, and working memory task with tACS data segments could be differentiated with accuracies ranging from 65-94%, far exceeding chance levels. EEG from the nBack and backwards digital recall tasks could be separated during stimulation, with an accuracy exceeding 72%. If residual tACS artefacts remained after the EEG cleaning these did not dominate the classification process. Significance: This helps in building confidence that true EEG information is present after artefact removal. Our methodology presents a new approach to validating tACS artefact removal approaches.