Awhile ago I wrote an Open Letter to Mint.com laying out some major concerns I have with their service and their security implementation. Almost all comments both here and on Hacker News and Reddit were divided into three categories:
- From non-Americans: How is a service like Mint.com even possible or legal? US Banks don’t have two factor security?
- I totally agree that Mint.com and their service is insecure and I don’t use them!
- I agree that Mint.com needs better security, but their service is great and anyway, it would be too time consuming/too expensive/too hard/too impractical to implement these security improvements.
Between the time I wrote that letter and now, we’ve seen RSA (the only major token based two factor security provider) have all of its hardware tokens compromised to much public uproar. At Sentry Data Systems, we’ve had two factor security implemented for years using time based cookies and additional security questions to challenge users when they were logging in from a device that hadn’t been previously authorized. This is similar to how many banks in the US do two factor security if they choose to implement it. While not a HIPAA requirement, we felt that it was a great feature to offer that provided an additional layer of protection. We’d originally offered RSA SecurID tokens to customers but found that most customers balked at the price, and even if they did use the tokens, many would simply tape it to their computer monitor or keyboard, or they’d forget the token at home which would cause quite the contentious support call. This experience brought to the forefront several issues that I had with hardware based tokens:
- Casual users or those who didn’t value the two factor security benefit would simply leave the token lying around or affix it somewhere – it wasn’t natural to expect a user to carry one more thing with them day-to-day.
- If there was a compromise, you have to replace all of your hardware, for everyone, everywhere.
- They were expensive.
- They were highly recognizable and screamed to informed observers that you had access to a system that was considered high-value by someone.
I even went so far as to start sketching out an iPhone app that we could deploy for our customers but it seemed like quite a lift to do it well (a correct implementation is key in cryptography systems) and it was with much delight that I ran across an outfit called DuoSecurity based in Michigan. They have really put together a fantastic service that provides both SMS based (challenge/response) and one-time password (via an iPhone or Android app) options for two factor security. I signed up for the service, installed their package on my Ubuntu Linux server, and within about 15 minutes, I had a very strong two factor solution that avoids all the drawbacks of the hardware token approach…for free. Yes – they provide up to 10 users for free to let you get your feet wet and see how the system works. With the token being my phone, I’m not going to forget it, it doesn’t draw attention to itself, I can’t tape it to my workstation, and they can update the software if they need to. If their service goes down, you can configure it to not require the second factor (the default) or you can choose to prevent logins and keep a private key around for last-ditch logins. Of course, for those of us running cloud based servers, there is still the risk that your hosting account could get compromised giving an attacker shell based access to the machine – hopefully Slicehost and other services will implement this type of additional security soon (Amazon’s EC2 cloud already implements two factor security as an option). Duosecurity can be easily implemented with any web application, a lot of VPNs, and on your Unix/Linux servers quickly and easily. If you’re doing anything with medical, financial, or other sensitive data you should definitely check them out. If you just like additional protection for your own servers and services, they’re a great option as well. Just in case you’re curious: Duosecurity put up a great blog post about the steps they’ve taken to prevent compromise if they came under the same attack as RSA. A few thoughts on improvement:
- Give me an apt package please! I don’t want to compile things, and I don’t want to edit configuration files. These things make it hard to deploy on lots of servers. I talked with a support rep from Duosecurity and they told me this is in the works already.
- Put a login form on your website! They email the login URL to you but I shouldn’t have to remember it.
- It’s a little unclear to me if the pricing scales well- if I’ve got the same 35 users access 100 machines, does that mean I pay 35x100x$3? That seems expensive. Course, it’s still way cheaper than RSA but at least you could bind an account to a token and not worry how many servers you were accessing. It’s possible that a single user crosses the server boundary, but again, I’m unclear on that.
Bringing it all back to the original point – there is simply no excuse why a service like Mint.com doesn’t use Duosecurity to protect its own user’s logins. But the second issue still exists – how do banks provide consumers of financial data access without compromising the entire account? A poor man’s solution of sorts could be taken by banks providing read-only accounts for customers that use generated, revokable passwords. Google takes this approach with its own two factor implementation for Gmail. You get texted when logging in normally, but for other applications, you generate a password that can be revoked at any point. It seems like a decent compromise – you can’t control the account from that login, and the password is of sufficient length and complexity that it’s unlikely to be brute forced. My initial suggestion of using Oauth is essentially the same thing. Congratulations to the guys/gals at Duo Security on providing a really great set of tools for developers and users. I really hope it catches on and more and more providers begin offering two factor as an option.