This generally happens when a virus from another animal (and thus with a slightly different version of the receptor) adapts to humans.
Basically, a gradient descent from a very different starting point ending up in a different local minima.
The question is, is the spike protein at a local minima (very probably quite close to it), and could it jump to another local minima then? Most likely not.
It's possible it will evolve to make our antibodies slightly less effective, likely as a trade off for infectiousness, but incredibly unlikely the spike protein would evolve to be unrecognizable to our immune systems. Historically, this doesn't happen all that often, except for viruses with chronic infections over years that can do a deeper search over the gradient, but even then it generally takes years to decades.
Generally though, within the context of machine learning, one of the benefits of gradient descent, especially when stochastic, is that we can get past those local minima humps. Does this hold less true with respect to the process of sequence mutation that viruses go through?