Auditors and public authorities may need to check how a company decided whether a partner was trustworthy. When AI is involved, what kind of explanation should the company be legally required to provide: a general description of how the system works, or a specific explanation for each individual decision?