A covert channel is an unintended path through which information leaks, such as error messages or timing. The system was not designed to share the information that way, but differences in what it does, how long it takes or what it reports can reveal facts to someone who is not supposed to know them.
It matters in shared systems. In a multi-tenant service, for example, an error that says a record already exists, or a response that is slower for data that is present, could let one tenant learn something about another, even though no data is returned directly.
Covert channels are often subtle and hard to find, and they may reveal only small pieces of information. Eliminating every one is rarely possible, so the aim is to reduce them where the information is sensitive. How serious a channel is depends on what it reveals and how easily it can be used.
To limit them, return the same kind of response whether or not a hidden record exists, avoid error messages that reveal details about other tenants, and include such leaks in your security testing and design reviews.