The way I understand it, if a client-side script running on a page from foo.com wants to request data from bar.com, in the request it must specify the header Origin: http://foo.com, and bar must respond with Access-Control-Allow-Origin: http://foo.com.
What is there to stop malicious code from the site roh.com from simply spoofing the header Origin: http://foo.com to request pages from bar?
