Is this interpretation of prediction intervals correct? I have a regression model relating to a ''normal'' population. If a new observation which corresponds to point $x$ is outside the prediction interval at $x$, do I have a reason to suspect the new observation is not in the normal population?
 A: A $100(1-\alpha\ )$% prediction interval should contain the next observation $100(1-\alpha\ )$% of the time. That is, it will not contain the next observation about $100(\alpha\ )$% of the time.
A: under some assumptions you constructed confidence interval with parameter $\alpha$ and it means that past observaions landed inside this interval with probability $1-\alpha$ and with probability $\alpha$ outside interval, that is all you can state
in your case this x is an unusual observation that lands outside interval but it's from same model (same distribution) 
you probably wonder if it's possible to construct interval that will "catch" all future observations - answer is NO since support of distribution covers whole observation space so confidence interval with $\alpha=0$ would cover whole space $(-\infty, \infty)$
in other words confidence interval allows you to define what was an outlier/unusual/low-probability observation in your experiment, but you can not state that "this observation is impossible" under your current model that uses normal distribution
