It's a neat way to exploit key rotation: clean now, malicious tomorrow.
Naturally, from a format perspective, it was always doable with any form of polyglots, as both formats typically don't overlap.
However, it's even more powerful: when one format is decrypted, the other format is garbled, which bypasses polyglot blacklisting (easy PDF polyglots).
Also, it's possible to bruteforce nonce to get some overlapping bytes, which enables polyglots unique to crypto-polyglots, such as PDF-PE (get the PDF you want now, it becomes Wannacry later).
When people ask me about the French language,
I tell them about "99 birds".
Why... what could go wrong?...
We read 99 as 'quatre vingt dix neuf' (4 20 10 9),
because 4*20=80 and 80+10+9 = 99 ;)
yes, '4 20' is the official way in France to read "80".
In other countries, they sometimes say "80" huitante (8 = huit),
which makes sense since 60 is soixante, even in France (6 = six).
Seriously, my son thought I was joking when I told him about 99 for the first time.
Especially because it requires to understand multiplication, which is a lot more complex than adding 1 or 10 for him at the time...
Another trick with `file` is that the types are checked by category, so some lesser known types will be checked before the classic exploitable ones (and not in the offset order either)