According to the documentation the ARIMA_PLUS model in BigQuery ML produces both confidence intervals as well as prediction intervals when forecasting. Upon trying it out I noticed the lower and upper bounds of the prediction interval are the same as the confidence interval. My understanding is that the prediction interval is always wider than the confidence interval for a given level, unless the error is zero.

This is also the case in their own tutorial (see table under step six):
arima single time series forecasting tutorial

I have fitted two instances of ARIMA_PLUS, both according to the linked tutorial, as well on another dataset. In both cases the bounds of the intervals are the same. I.e., conf_lo = pred_lo and conf_hi = pred_hi. My expectation is that they would differ, and the prediction interval would be wider.