The insecurity and usability problems of passwords are well known. Passwords are something you know, so they are vulnerable to forgetting, which happens often. Thus, email is widely used as a backup to reset passwords.
Email has a lot of problems too. Similar to passwords, people typically reuse the same email address across lots of services, creating a privacy risk if the email is discovered from the service. It is increasingly difficult to get an email address without giving personally identifiable information to the email server. As a deterrent against spam (and probably also because it makes it easier to target ads at users), free email services typically require providing a phone number which is easy to associate with a particular person. Paid email services might not require a phone number, but paying for a service without personally identifiable information is difficult as well, and relying on paid email service subscription is vulnerable to changes in financial circumstances. Also, it’s difficult to reliably self host an email server today. In addition to the privacy issues, the centralization created by reusing an email account across many services also creates a security risk because a compromise of the email account would compromise lots of other accounts.
Nowadays, we don’t need passwords nor emails to register or authenticate to a service. Discourse already supports FIDO and TOTP, but it still requires a password and email address to register and authenticate. It would be great if Discourse made passwords and emails optional in favor of FIDO and TOTP.
One factor authentication with FIDO can be really convenient, but it is vulnerable to loss or destruction of the single FIDO token, similar to the issue of registering with a password but no email address. To resolve this, I propose that users would be required to provide at least two factors to register, which could be any combination of FIDO, TOTP, and/or password. Users who want emailless & passwordless authentication could simply register two FIDO roaming authenticators like Yubikeys. Users could be advised (or potentially required, especially for administrators) to register more than the minimum of two factors to avoid losing access to their accounts.
As FIDO platform authenticators are being built into more and more devices these days with Windows Hello, Apple Touch & Face ID, and Android, this email-less registration system could be usable by nontechnical users who do not own specialized roaming authenticator hardware like a Yubikey. Users could register with the FIDO platform authenticator plus a password. One factor authentication with the FIDO platform authenticator could work seamlessly with such a setup. However, this would create a usability problem for authentication on new devices because users wouldn’t have the FIDO platform authenticator available on a new device and relying solely on the password to setup a new device wouldn’t be secure. To resolve this, I propose a workflow similar to how Matrix authenticates new clients. The user could try to login on a new device with that device’s FIDO platform authenticator (a new factor) and their password (an already registered factor). This would not actually log in, but it would create a request to approve the new FIDO authenticator in the account. The UI on the new device would then direct the user to log in on a device they already have registered to approve the new device. With FIDO platform authenticators built into mobile devices, this could be practically usable for secure authentication without specialized roaming authenticator hardware or sacrificing the ability to use any ad-hoc device like a public kiosk.
I just came up with this anonymous registration & authentication system yesterday after receiving my Yubikeys. I am not aware of any systems which implement this. I would love to see a mature and already widely deployed web application such as Discourse pioneer a future without email or other personally identifiable information being required to use the Internet.
）仅使用 Google SSO，而 Discourse 已经提供了该选项。