This has definitely happened in endgames, e.g.:
Queen and a minor piece versus two rooks: this is usually a draw for a knight and a win for a bishop, although the win takes up to eighty-five moves. The best method of defense is to double the rooks on the third rank with the opposing king on the other side and keep the king behind the rooks. This case with a bishop and queen versus rooks is unusual in that such a small material advantage forces a win. It was thought to be a draw by human analysis, but computer analysis revealed a long forced win (Müller & Lamprecht 2001:404), (Nunn 2002a:328–29,367,372).
It also happens earlier during the middlegame, although this will depend on the time the human spends analyzing. It's a lot easier for the human to find the winning line if they know there's a way to "break the fortress".
Example (39 minutes in):
[FEN "1q4rr/5p2/kpb2b2/p1p1pP2/P1PpP1pN/1P1P2P1/4QBN1/1K4R1 w - - 0 1"]
(Black to play)
Black is up the exchange, but the position is completely locked. Black could sacrifice an exchange on h4, but White has the g4-pawn well-covered and trades will only help White draw, since Black's c6-bishop is never penetrating White's position and it'd be an OCB endgame as well. Draw?
In fairness to Nakamura, he didn't actually say it was a draw (he was playing against Komodo, after all), but you can see his reaction to Komodo's next move.