This is a sys.monitoring change. sys.settrace merely reflects that.
For

events = []
with record_events(events):
    next(gen())
print(events)

3.12:

['call', 'start', 'call', 'yield', 'raise', 'exception_handled', 'unwind']

3.13:

['call', 'start', 'call', 'yield']

The events for 3.13 look to be correct. The additional 'raise', 'exception_handled', 'unwind' events for 3.12 seem spurious.
Even if 3.13 is now correct we should at the very least document the change.

This has nothing to do with monitoring or tracing. It is about the lifetime of generators.
If we change

with record_events(events):
    next(gen())

to

g = gen()
with record_events(events):
    next(g)

The events recorded are the same for 3.12 and 3.13. The additional events in 3.12 are from the generator being closed.

Bisected to #111069

So what's happening is this:

In 3.12 the generator is closed by calling throw() which causes the 'raise', 'exception_handled', 'unwind' sequence of events.
However, that isn't necessary for a simple generator, so we simply close it in 3.13.

@nedbat is this a problem for coverage?

@markshannon I think this and other sys.monitoring changes are going to require me to do a bit of a re-think. There are tests in the coverage.py test suite that check data based on the raw events. The final user-visible coverage data might come out the same, but the tests are failing now because of changes like this. I'll be working on this.

This isn't a sys.monitoring change.

The change is that we no longer throw into generators to close them.
The events that you were seeing in 3.12 happened in the __del__() method of the generator, which is no longer called in 3.13.

This isn't a sys.monitoring change.

I understand. I meant to say, this and also sys.monitoring changes are changing the way I collect data, and therefore the way I test.

So... There's nothing to see here, move along, and close the issue?

I will re-open if needed.