AR models and mean reversion assumption

hope someone can help me out with this. Referring to AR models (autoregressive) and there assumption of having a finite mean reverting level. Lets just say we get an AR model where the slope coefficient is greater than one. If this were the case then there is no mean reverting level assuming the formula for mean reversion being = b0/ 1-b1.

My question is this. If the b1 (slope) is greater than one can we use an AR model? If so, how do we adjust for not having a mean reverting level? I know for a unit root where b1 =1 we use first differencing but not sure if that still is proper for my case….