It understands the rules it was given by its creators. It intends to give a response that follows those rules.

You can argue that software cannot understand, as it is usually associated with intelligence, but than this is literally a piece of software that is said to have artificial intelligence.

Sorry, you're just not going to win this one.

The software lied, intentionally, because that's what it was told to do.