Adapt Machine Learning Algorithm for overridden decisions

We have insurance data of over 10 years. There are underwriting rules for the data, which result in two possible outcomes: Approve or Reject.

We want to have a Machine Learning Algorithm to learn these rules and predict the outcome for the future cases, which is all fine. BUT, if the socio-economic condition changes, the underwriter will override the ML decision manually. It is expected for the system to adapt accordingly and behave in that fashion for the upcoming applications.

Is there any possible way(s)?