I'm intrigued now, I hope someone can give details!
For example, with Facebook, once a user has logged in and given Application X permission to access their details, FB will send the user's id, name, etc. (whatever data the user has granted access to) along with a unique access token. The next time, and every subsequent time Application X wants to access the FB API on the user's behalf it is required to send that access token. From javascript you might be able to change the token, however, Application X's next interaction with the FB API will fail if the token is invalid and there is no way to derive a token value from a FB user's id.
I don't have to change the token. I just have to change the data given by facebook (including the uid) before the website's dumb javascript uses it in a post back to the server. Since it's not signed by Facebook, how can the website's server trust the uid? Never trust your user input.
Facebook returns a uid. When the user takes an action, this uid is sent to the server. The server trusts the uid, and saves this action as taken by the user identified by this uid.
And let's say it's not the uid. Let's say it's the user's name.
It trusts the user input basically. But it should probably be getting it directly from facebook, or in a signed structure, right?