Caro Kann opening

"Is the Caro Kann Advance variation winning for White?"

Not proven so far.

Engines and centaurs (engine plus human) are in continuous improvement so the last word on a particular opening variation is yet to be uttered.

You don't give a link to the game nor the particular configurations of the engines.

For the tournament to find the strongest engine a test bed is set up, two engines play in the finals, 100 games where they in turn take the white and black side of the same opening variations (a human picks the variations to check and hopefully provide a good contest).

The most recent championship was won by Stockfish over Leela and the 100 game match could be found on chess24 site. You could check the games where the Caro Kann was played, I don't know off hand if the variation you are interested in was played.

When engines rated 1000 and more Elo above Magnus Carlsen sometimes the moves played are quite mysterious, well more than sometimes!

I have recently analysed the Short variation of the Advance Variation quite thoroughly with Leela and heavily with Sockfish, and I can say that I have not found a forced win at all.

There are several ways to a draw for Black. Overall, lines with an early ...c5 performed better.

I don't know the game you're referring to, but I doubt it shows, all by itself, that the Caro-Kann Advance is winning for White (and therefore the Caro-Kann itself is losing for Black.) Surely Black could have varied somewhere before getting down to tablebases.

The fact that an engine will play the opening at all means it is programmed into the engine's opening book - which means that, at least, the makers of that opening book think it's worth playing as Black.

One database shows that Black wins in the Caro-Kann Advance 29.7% of the time, and White wins 42.5% of the time. That's slightly worse for Black than average, but White still cannot manage a win in half the games, so therefore I don't think you can say it's "winning" for White. If it were winning, White would win!