Most apps have usage reporting enabled either permanently, or by default. It may be just in case of a crash, or it can constantly be collecting and reporting every tap and scroll you make, how long you hover before scrolling, if your screen is on, if you’re not switching to other apps, stuff like that.
Furthermore, the OS is constantly aware of other information, like from the motion sensors, wifi signal strength, connected devices, storage space, installed running apps, battery etc. It collects that data and can pass it on to the apps as well.
So all that information is then packaged together with some unique identifiers, phone number etc.
So there isn’t much need for camera to be always recording, since the system always knows where you are and what are you doing. Combine that with data of other users, and they know who you are with and what you are doing.
As for microphone, depends. Voice assistants do listen all the time for keywords and can probably be made to record everything, but that doesn’t seem to be the case. But they can still record if they mishear a command.
Worth noting that this kind of telemetry isn’t programmed by every app maker from scratch, but is done by libraries that people insert into their apps. Such as those by Google or MS, but alao others. So the library creator has access to all the data from other apps too, and can share it with customers and other parties.