tbh refactoring (if it can be done safely) could be a huge help to orgs and fantastic AI use case. Rarely do businesses want to spend the resources to improve a codebase. Hope it goes well for IBM.
How can you verify that it's actually running safely? If you'd have a comprehensive testing framework in place, you wouldn't need AI driven refactoring in the first place.
You could code review the commits it makes I suppose. Saves you having to think of how to refactor something but still makes it fairly easy to mitigate the AI doing something terribly wrong.