Ask for advice, it does exist.
Compliance
Today many regulators including the European Union has issued guidelines for working with AI, these condensed focus on three things, ethical, legal and technical robust. The legal part has a large focus on GDPR and the rights for each individuals in regards to data and sharing, storing personal information. Without going into detail, it is important to comply to these rules, but you might also say this law has nothing to do with AI, it is just that AI lives on data therefore the focus and importance.
Liability
Not many cases of AI, where things has gone wrong exist. This is a little bit of a problem today, as autonomous cars, vehicles etc in general does not exist for real. These are still in the R&D department, and they won’t hit the roads until this has been resolved. Right now the simple rule to think of is. If you own a dog and despite you believe it is somewhat intelligent is bites your neighbour, you will face charges. You are liable for your dog. This could for instance be the case of autonomous customer interaction and your model is biased not granting loanes to a specific group of people due to religion or something else that would be considered to be discrimenary. If you are Head of Customers Loans, then you can expect problems.
Explainable
Regardless of what function or solution based on AI, it is expected that you or your team can explain the outcomes of it. Results are not created randomly, neither are those outcomes or functions it provides. There can in these processes of execution be several sources to failure of flaws. This could for instance be a biased algorithm, or biased data. It is in general easier to adjust a biased algorithm than biased data. The need for openness and easily dissected models and data used as either training data or live feed data is important. Therefore is it also important to carefully document and monitor changes to models and data, as with all other IT-development.