Suppose some large private company has to decide whether they're going to build a new facility in city A or city B. This is useful information for all kinds of reasons. If you're a vendor then you need to start making preparations to set up shop in the city where your big customer is moving etc.
The company's analysis shows it would derive a $10M advantage from building in city A. The prediction market is correctly leaning that way. If there are only enough counterparties that someone who now bets on city B and wins would make $5M, everything works the way it should and the company goes with city A. But if there are enough counterparties that a winning bet on city B would net you $25M then the company can place the bet, eat the $10M loss by choosing city B and come out $15M ahead.
But the $10M number isn't public. It's essentially the thing you wanted the market to predict and it could be arbitrarily larger or smaller than that. So how are you supposed to know if the prediction market will be predicting the result or determining it?
On the other hand, at the point where the prediction market winnings are material enough that they might alter the underlying decision itself, you've clearly got an antisocial structure. Prediction markets that don't want to be seen as mere prop betting venues should refuse to run markets on those questions.
It is insider trading, the thing everyone here is talking about