If an attacker can force you to sign arbitrary messages, that's certainly a security flaw. The solution is to take good care that you approve of whatever content you are thinking of signing whether it be cheques, loan agreements, wills or numbers.
So: the danger isn't revealing private keys (I mis-remembered), the danger is a naive developer will see the signmessage RPC command, not realize that signing arbitrary data can be dangerous, and put up a web page that lets somebody enter arbitrary data to be signed with one of the developer's public keys.
This might just be a documentation issue, although if signmessage was changed to sign a hash of the passed-in message instead of the message itself then it would be completely safe.