I had this happen as well. I essentially deployed a new witness, updated it to the matching ESXi version of the cluster using an image I exported from vLCM, then moved the cluster to it and everything was happy.
Not sure why it had the other drivers, but doing the re-deploy seemed to resolved the issue this time anyway. Will need to see if next time I have the issue, which then I will have to conclude vLCM is doing the deed and I will need to re-evaluate the whole image based remediation of my two-node clusters.