How accurate are the bot Elo ratings?

I'm a beginner, my rating bubbles around just under 600 at the moment. I know this is a fairly good barometer of my actual ability because, if I play someone 650+ I almost always lose. If I play someone 580 or below, I (usually) win and anyone in that 580-650 is usually a good battle.

However, I have beaten all the AI's in the "Beginner" category (up to 850) and usually beat them every single game, and have beaten the first 6 bots in the "Intermediate" category up to and including Sven (elo 1100) and win most, but not all, games.

So my question is - are the ratings of the AI characters a bit off, or is there some other conceivable reason why I can beat a 1100 computer but struggle with a 650 human. This is not a complaint about chess.com or anything like that. I am just intrigued by it. For the record, I guess I play about 50/50 games against the computer and real people.

Correct, the bot ratings are not representative to human players.  Perhaps around the 1800 mark the bots become more representative.

Someone I'm inclined to agree with said, "It's hard to make computers play badly but not too badly."

Interesting, thanks.  I guess it must be tough to make a computer play poor, but plausible, moves.

I was running a theoretical experiment of registering another account and playing rapid games against humans, but using the moves of a bot copying from another computer. Just to found out the realistic rapid chess.com ELO of that bot.
I won't do it but I wonder if someone did that.