For LUW, you can modify a derived table, but you need something that uniquely determines each row if you want it to be deterministic. On the other hand, if the rows are identical, it probably does not matter which one you modify. Whether this work for i-series I don't know, but you can try the following:
create table t (a int not null, b int not null);
insert into t (a,b) values (1,1),(1,1),(1,1),(2,2),(3,3),(3,3);
select t.*, row_number() over (partition by a,b) as rn from t;
update (
select t.*, row_number() over (partition by a,b) as rn from t
) set b = 9 where rn = 1;
delete from (
select t.*, row_number() over (partition by a,b) as rn from t
) where rn > 1;
select t.*, row_number() over (partition by a,b) as rn from t;
Fiddle