The Spam Filter with Morals

A company develops a new AI spam filter for emails.

It should not only detect ads and scams, but also filter out “toxic content.”

At first, everything seems to work well.

Less spam, fewer insults, less stress in the inbox.

But after a few weeks, something strange happens:

Harmless messages suddenly disappear as well.

Complaints, criticism, and very direct wording are especially affected.

Question:
Why does a “moral” spam filter suddenly remove normal and important messages too?

Solution will be unlocked tomorrow.