I can confirm that this is indeed a problem on the 33522A generator. It stays at half-level for about 3.5 msec when changing many of the waveform parameters. It has recent firmware version 5.02-1.19-2.00-58-00. Perhaps someone can check 33522B generator, although I am not sure if there is any difference between them. The problem is present when using both square wave and pulse waveform. It reminds me of a similar transient issue that appeared at first on Rigol DG800 generator and also blew up someone's circuit, but Rigol quickly fixed it in the firmware.
Keysight 33522 does not specify a separate "idle value", so the voltage between bursts is not so well defined, but defaults to 1st point of the waveform. When using sine waveform one can set any idle voltage by adjusting the starting phase and when using square waveform one can switch between high and low with starting phase. However, when reprogramming the waveform it goes to 50% point, not to the 1st point of the waveform, I think its just a bug.
In contrast, Rigol DG800 has a separate specification for idle value that can be set as top, bottom or any point of the waveform. And when changing parameters it stays at that idle value.
I would encourage you to contact Keysight directly and demand an explanation. It's surprising that this bug had not been previously reported, but looking back I think it might have caused some unexplained damages in our lab as well.