I have registered and stacked subs in DSS and PixInsight and see the same thing. I saw this same thing once in a set of subs I processed in DSS. I tried PixInsight with the same subs and it was gone.
I assume the culprit is some setting I am using. I just ran PI preprocessing, stopping at registration and there are no defects. I am working with 13 90-second subs. The PI integration setting I used are:
Pixel Rejection Algorithm: Windorized Sigma Clipping
Sigma Low/High: 4.0/2.0
Scale Estimator: Iterative k-sigma/biweight midvariance
Normalization: Additive with Scaling
Weights: Noise Evaluation