missing column name restricted by a sub-select results to wrong dataset instead of error

Trying to follow an obscure statement, I created another one that provided weird incorrect result. Simplified variant looks like:

with 

xxx as (

  select 2 /*as id*/, 'row 2'

  union all

  select 3, 'row 3'

),

tab as (

  select 1 as id, 'row 1' as txt

  union all

  select * from xxx

)

select * from tab

where id in (select id from xxx)}}

that "mistakenly" resulting to:

id	txt

1	row 1

2	row 2

3	row 3

however initially expected this (row 1 was unexpected):

id	txt

2	row 2

3	row 3

however after attentive view the expectation is an error:
Unknown column 'id' in 'field list'
(in sub-select, because xxx really misses it).

Although the reason for that was a "mistake" from my side: forgotten the commented part as id that provides a column id to the xxx-statement, the wrong result is totally unexpected here.

By the way it looks like the union-ing in the second with-statement tab plays an important role by the issue, because somehow it seems to "follow" the column name id from this second statement (after union) in a sub-select to the first one (just doing it wrong).

If one rewrite it like this (with a column col instead of id which then completely missed in with-statements:

...

- where id in (select id from xxx)}}

+ where id in (select col from xxx)}}

it would fail as expected with an error:
Unknown column 'col' in 'field list'

So a conjunction to id by a with-statement + union seems to arise there internally.