I experimented a bit more and found the following:

If I rewrite the HQL to this (don’t select id, use distinct instead of group by):

insert into TabData (col1, col2, col3)
select distinct col1, col2, col3
from TabDataStaging

Hibernate 6 does the following:

insert into HTE_tab_data(col1, col2, col3, rn_)
    (select distinct p2_0.col1,
                     p2_0.col2,
                     p2_0.col3,
                     row_number() over () -- adding this column effectively disables the `distinct`
     from tab_data_staging p2_0);

Followed by the updates on the HTE table you were suspecting earlier:

update HTE_tab_data set id=? where rn_=?
update HTE_tab_data set id=? where rn_=?
...

Followed by

insert into tab_data(col1,col2,col3,id)
select p1_0.col1,p1_0.col2,p1_0.col3,p1_0.id
from HTE_tab_data p1_0

The final insert into tab_data can then fail due to unique constraint violation (I have a composite unique constraint on col1, col2, col3), because even though I’m using distinct, the row_number() clause added by Hibernate effectively disabled it…