My understanding has been that, when behind a firewall and with proper security measures to protect one from hacking attempts, autofill is safe. I'll leave it to others to hash this out.
This is incorrect info as an unauthenticated remote attacker could exploit auto fill by constructing a malicious page under a domain that matches stored info, i.e. strong keychain password will not protect you at all. This is a security issue common to other browsers.
This is true, but I have never heard of a phishing attempt that does this. It can of course theoretically be done, but it would require hacking into the web site in question to create pages on that domain. In other words, the people doing the phishing would have to hack into, for example, bankofamerica.com to create the page where they then retrieve your autofilled login information, before Bank of America discovers that they've done so and stops them retrieving the information sent from that page.
Message was edited by: Rachel R