Is it possible for an AR(1) model to have a unit root after first differencing and taking the natural log?
Yep. In that case, you apply a second differencing. The challenge now is to properly interpret the results. Further differencing is a nonsense because interpretation on the results becomes unintelligible.