Seconds, and the players themselves, absolutely use engines to help them find moves, but you cannot just let an engine run and think that it will spit out useful things. There would be too much information, and most of it useless.
First, nowadays, more and more, elite players are resorting to using Leela and Fat Fritz (neural network type programs) since they tend to be better at finding better opening moves. Stockfish is still the king when it comes to brute-force analysis in the middlegame.
Second, if you assume that the game is relatively equal from the opening, just letting a computer run theoretically should not find anything except a lot of equal lines since it will only analyze best moves for both sides. You need human intervention.
That is where the second comes in. Most seconds are very strong GMs with a great depth of theoretical opening knowledge and understanding, and they look at a position, and come up with an idea that they want to test in the player's main lines, and they basically play "advanced chess", combining their great human understanding with the computer's immense analytical ability. Of course, it is not impossible that they will stumble upon a clearly winning line, but at that level, both sides have thoroughly checked their opening repertoire with computers, and the players have such a strong positional understanding, that it is highly unlikely they would play something that weak in the first place, computer-checked, or not. They are simply looking for new move that leads small advantages that lead to an uncomfortable game for the opponent...a game they can press for a long time, hoping that it will create an unforced error.