Basically how reducible is personal information until it's no longer personal information?
When it can no longer be used, alone or combined with other data, to identify a person (or small number of people).
Context matters, if you have data on the people in say New York City, (pop. 8 million+) there are only 676 2 letter initials which averages out to about 12,000 each BUT initials are not uniformly distributed: there are a lot more JS out there than XX. In fact XX could be so rare that you can identify an individual in NYC just from this.
Of course, if you are using all initials then in, English speaking countries most people have 3 names and so, 3 initials; this not only reduces the number of people in each “bucket” it makes it way easier to identify unique individuals with 2, 4, 5 or more names.
If your data set is smaller than this and people have a way of working out who was in your data, then the prospect of being able to identify an individual just from their initials is much greater. For example, my family has 4 individuals and there are 4 unique first initials so in that group, the initial clearly identifies each of us.