I should have said that the principle in question applies for all methods of communication other than some of those which rely on a challenge from the server.

Some challenge-response systems won’t be safe against such an attack either, if you don’t have public key encryption and you have the same algorithm and data set used by all servers then any compromised server should be able to gain credentials for others. The problem is that doing this properly requires more compute power on the client side – which can be a problem for a smart-card or other limited device.

Maxim: I’ve replaced the “spambot” message with one that makes things easier for humans.

Barak: It’s not a OTP. If you have server A and server B accepting the same login credentials then if you login to server A which has been compromised then that server could instead of authenticating you use the credentials you supplied to login to server B. This applies for all methods of authentication. If server A authenticates you and if the authentication server has not been compromised then server B will not accept a login with the same code from the Yubikey.

Sure everything the Yubikey does could be done with some software and a USB storage device. The point of the Yubikey is that it’s small enough and compatible enough that you can carry it everywhere.

Maxim: If you are accused of being a “spambot” then you missed the arithmetic question. I don’t like the way it handles errors in that regard and plan to change it.

Barak: pam_usb does that, sort of. That project has pretty slow development though. I even think pam_usb got dropped from Fedora for it. The limitation of such a setup would be though, that it would not work – or at least: not easily – over a network. Yubikeys are designed to work over the network. Mitm attacks are an issue for a lot of other network authentication mechanisms too. I a bit short on time, but I invite you to read a bit in the docs on the Yubico website. It’s not as simple as you make it sound.

This sounds a bit vulnerable to me.

(Obviously two-factor authentication, e.g., an additional per-server password, would make such attacks trickier, although perhaps not impossible. Additional passwords, however, also make the device itself of much less utility, since the password alone provides reasonable security.)

Also, from a practical perspective, couldn’t similar functionality be provided by a USB storage dongle holding some secret information (long one-time pad) together with some program to trickle it out on demand in the right way?