Heteroskedasticity
Non-constant error variance, so A4 fails. OLS stays unbiased, but the usual standard-error formula is biased, making t- and F-tests invalid, and by Gauss-Markov Theorem OLS is no longer efficient.
When to use
In R output only the coefficient estimates remain trustworthy. Fix it with Weighted Least Squares when you know the variance form, or with the White Correction when you do not.
See Lecture 12.