Most likely you already know about the theory of the colliding miners planning a 51% attack to steal funds of transactions sitting on segwit addresses. Let's say it actually happens: Jihan and co gather and enough hashrate collides to steal segwit funds. Let's say the funds are a small amount and the address has been inactive a lot of time. Most likely the owner wouldn't even notice, or by the time he notices, nobody would believe that it actually happened, unless you can answer the question of the thread:
In the event that it actually happened, would there be any way to prove it, or it would be seen just as a regularly sent transaction?
Since the miners don't have the correct private key, they will be unable to provide the correct signature for the transaction that they use to "steal the funds". Therefore, everyone else on the network will reject their invalid block which contains the invalid transaction. They will have wasted their time and money and they will have accomplished nothing.
I think OP is alluring to the anyone-can-spend "vulnerability" that has been a talking point against the SegWit softfork for a while. What usually got ignored during this discourse was that "exploiting" this attribute of SegWit transactions would require a hardfork, basically rolling back BTC's SegWit upgrade and creating a shittier version of BCH.
So in theory, at least the way I understand it, anyone -- not just miners -- can try to spend inputs from a SegWit transaction while ignoring the correct private key. But while legacy (ie. non-SegWit) nodes would accept such a transaction, the majority of the network -- and thus the canonical BTC blockchain -- would reject it.