Yet another blog post that starts with a tweet, or two. Even though I’m not by either trade or training clearly a security person, I have a vested interest in security as many of you know, and I have written (or ranted) a lot about passwords and security in the past. I have in particular argued for LastPass as a password manager, and at the same time for U2F security keys for additional security (as you can notice in this post as I added an Amazon link to one).
The other week, a side-by-side of the real PayPal login window and a phishing one that looked way too similar was posted, and I quoted the tweet asking PayPal why do they still not support U2F — technically PayPal does support 2-factors authentication, but that as far as I know that is only for US customers, and it is based on a physical OTP token.
After I tweeted that, someone who I would have expected to understand the problem better, argued that password managers and random passwords completely solve the phishing problem. I absolutely disagree and I started arguing it on Twitter, but the 140 characters limit makes it very difficult to provide good information, and particularly to refer to it later on. So here is the post.
The argument: you can’t phish a password you don’t know, and using password managers, you don’t have to (or have a way to) know that password. Works well in theory, but let’s see how that does not actually work.
Password managers allow you to generate a random, complex password (well, as long as the websites allow you to), and thanks to features such as autofill, you never actually get to see the password. This is good.
Unfortunately, there are plenty of cases in which you need to either see, read, or copy to clipboard the password. Even LastPass, which has, in my opinion, a well defined way to deal with “equivalent domains”, is not perfect: not all Amazon websites are grouped together, for instance. While they do provide an easy way to add more domains to the list of equivalency, it does mean I have about 30 of them customised for my own account right now.
What this means is that users are actually trained to the idea that sometimes the autofill won’t work because the domain is just not in the right list. And even when it is, sometimes the form has changed, and autofill just does not work. I have seen plenty of those situations myself. And so, even though you may not know the password, phishing works if it convinces you that the reason why autofill is not working is not because the site is malicious, but just because the password manager is broken/not working as intended/whatever else.
This becomes even more likely when you’re using one of the more “open” password managers. Many think LastPass is bad, not just because they store your password server-side (encrypted) but also because it interacts directly with the browser. After all, the whole point of the LostPass vulnerability was that UI is hard. So the replacements for geeks who want to be even more secure usually is a separate app that requires you to copy-paste your password from it to the browser. And in that case, you may not know the password, but you can still be phished.
If you want to make the situation even more ridiculous, go back to read my musings on bank security. My current bank (and one of my former ones) explicitly disallow you from using either autofill or copy-paste. Worse they ask you to fill in parts of your password, e.g. “the 2nd, 4th, 12th character of your password” — so you either end up having to use a mnemonic password or you have to look at your password manager and count. And that is very easily phisable. Should I point out that they insist that this is done to reduce chances of phishing?
I have proposed some time ago a more secure way to handle equivalent domains, in which websites can feed back information to the password managers on who’s what. There is some support for things like this on iOS at least, where I think the website can declare which apps are allowed to take over links to themselves. But as far as I know, even now that SmartLock is a thing, Chrome/Android do not support anything like that. Nor does LastPass. I’m sad.
Let me have even more fun about password managers, U2F, and feel okay with having a huge Amazon link at the top of this post:
from new evidence we were able to observe how NilePhish are targeting two-factor codes to get access to Gmail accounts of activists: pic.twitter.com/OrNTa6Qh40
— Ramy Raoof (@RamyRaoof) February 23, 2017
Edit: full report is available for those who just don’t believe the infographic.
This is not news, I have written about this over two years ago after U2F keys were announced publicly — I have been using one before that, that is not mystery clearly. Indeed, unlike autofill and copy-paste, U2F identification does not involve interaction with the user-person, but is rather tied to the website’s own identification, which means it can’t be phished.
Phishing, as described in the NilePhish case above, applies just as equally to SMS, Authenticator apps and push-notifications, because what the user may perceive is just a little bit more delay. It’s not impossible, though a bit complicated, for a “sophisticated” (or well-motivated) attacker to just render the same exact page as the original site, defeating all those security theatre measure such as login-request IDs, custom avatars and custom phrases — or systems that ask you the 4th, 9th and 15th characters of your password.
User certificates, whether as a file enrolled in the operating system or on a smartcard, are of course an even stronger protection than U2F, but having used them, they are not really that friendly. That’s a topic for another day, though.