Autism is a neurodevelopmental condition, which means differences in the brain are expected to be underlying the symptoms. Research into functional connectivity has attempted to find these differences, but relies on assumptions of stability which have not been studied for autistic individuals. This study assessed the stability of functional connectomes across time and viewing condition. We used three scanning sessions on different days and three different viewing conditions to determine the stability of connectomes of autistic and non-autistic participants. To determine how this may affect clinical application of functional connectivity we also used Machine Learning to classify autistic and non-autistic participants. We found no group differences in stability and no effect of scanning day. There was an effect of viewing condition and of number of included runs. Connectomes had higher stability when including more runs and when using video conditions rather than the rest condition. This suggests previous assumptions of stability can be applied to autistic populations, but assumptions about stability must take the viewing condition into account. In the Machine Learning classification balanced accuracy outcomes were below chance level. The session and viewing condition that runs were taken from for classification had no significant effect on balanced accuracy. The number of runs included did have an effect, the balanced accuracy decreased when more runs were included. This may be a sign of overfitting and means that more stable connectomes do not necessarily lead to more accurate identification of differences between autistic and non-autistic participants.