I'm not an expert in seismometry, but I think there'd be quite a lot of challenges to overcome. Off the top of my head:
I don't know how precise iPhone accelerometers are, but modern seismometers measure on the nanometre scale.
There's also the issue of orientation. Seismometers measure X, Y and Z and are fixed in place, aligned with latitude and longitude or at an offset. It looks like you can get X, Y, Z from an iPhone accelerometer, but the phone could be in any number of different orientations. It would be necessary and difficult to rectify all those different orientations, as calculating magnitude on a horizontal or vertical component makes a difference.
Seismometers also have well known transfer functions, which are required because a step in calculating most magnitude types is simulating them as a particular type of seismometer, e.g. ml or local magnitude is taken after simulation as a Wood-Anderson seismometer.
Orientation wouldn't be that hard an issue to overcome, phones have magnetometers to act as a compass, and accelerometers which indicate the direction of gravity, so assuming you have a decent idea of where on earth the phone is located, you can work out its orientation in 3 space.
Precision can to some degree be made up for by quantity. If data from all the phones in an area skew the same way, it can stand out even if the signal would be impossibly lost in the noise from a single detector.
More problematic would be calibration. A phone at the top of a tall building vibrating like a tuning fork is going to have a radically different signature from one at the same altitude but in the pocket of a mountain climber, which in turn is going to be very different from one at sea level but close to a highway. If you had a sensor sitting in the same place for a long time, you could probably characterize it, but assuming these are phones people are going to pick up and move after a couple minutes I doubt you could do more than a very crude guess.
I don't know how precise iPhone accelerometers are, but modern seismometers measure on the nanometre scale.
There's also the issue of orientation. Seismometers measure X, Y and Z and are fixed in place, aligned with latitude and longitude or at an offset. It looks like you can get X, Y, Z from an iPhone accelerometer, but the phone could be in any number of different orientations. It would be necessary and difficult to rectify all those different orientations, as calculating magnitude on a horizontal or vertical component makes a difference.
Seismometers also have well known transfer functions, which are required because a step in calculating most magnitude types is simulating them as a particular type of seismometer, e.g. ml or local magnitude is taken after simulation as a Wood-Anderson seismometer.
Not saying it couldn't be done, just lots of problems to solve. There's always https://raspberryshake.org/!