Hacker News Reader

Comment by joe_the_user | Hacker News Reader

Third party logins really don't supply any enhanced security as such. But they do supply more "security theater" which is not actually what you want.

Three factor does provide more security but there's no reason to involve a third party - in fact, the more parties, the more likely someone is to fumble something.

View on HN

You're right. It's the poor man's identity federation.

Before I start, let me make myself very clear - I think Dan did an awesome job with Foamicate. Seriously. One guy. The balls to put something out there that's innovative and seeks to solve a REAL problem. I'm impressed. And if I were in a position to, I'd ask Dan to work for me.

That said, it's worth (IMO) considering what an identity system should provide -

User Control and Consent: An identity system should only reveal information identifying a user with the user's consent. The Facebook/Google/OpenId model fails this one.

Minimal Disclosure for a Constrained Use: The solution that discloses the least amount of identifying information and best limits its use is the most stable long-term solution. Again, OpenId falls short in that the identity provider knows where the user is going.

Justifiable Parties: Digital identity systems must be designed so the disclosure of identifying information is limited to parties having a necessary and justifiable place in a given identity relationship. Microsoft Passport is a great example of why this's a bad idea.

Directed Identity: A universal identity system must support both "omni-directional" identifiers for use by public entities and "unidirectional" identifiers for use by private entities, thus facilitating discovery while preventing unnecessary release of correlation handles. My Facebook identity is public, but I can't and shouldn't use that to access a private resource like a collaboration site on my company's extranet.

Pluralism of Operators and Technologies: A universal identity system must channel and enable the inter-working of multiple identity technologies run by multiple identity providers. The part where anything that's exclusively browser-based fails. The same system should work online, from a phone, and from a PC or server running any arbitrary OS.

Human Integration: The universal identity metasystem must define the human user to be a component of the distributed system integrated through unambiguous human-machine communication mechanisms offering protection against identity attacks. UX. The usability problems with Foamicator have been discussed here already.

Consistent Experience Across Contexts: The unifying identity metasystem must guarantee its users a simple, consistent experience while enabling separation of contexts through multiple operators and technologies. A smart card is a good example of this - I have a card for cash, a loyalty card for coffee, and an access card for work. Using them is consistent, event though their context is not.

The only technology I know of that meets all 7 laws is information cards (Higgins Project, and CardSpace). Information cards haven't taken off because adotpion by identity providers is, well, basically zero.

Which is a shame.

Anyway, if you're interested in this you can find more information at http://www.identityblog.com/.

Thank you for the compliments. With this I was hoping to spark conversation because I've been wondering why something similar hasn't been done yet. We use public key crypto to authenticate with SSH so why not also use it for websites especially with the recent leaks of information like what happened with youporn.

Maybe add one more:

Allow user to securely login to websites from virus infested machine in an internet cafe.

How do these various systems compare based on that law? Google one time password or card space?

Neither.

Google and identity is a bad idea regardless of the problem you're trying to solve (http://www.identityblog.com/?p=1204). It's the worst of the three.

OTP is good, but challenge-response is better. Is your OTP from a second factor?

Card Space is my personal favourite, but the [card] portability issue hasn't been solved. U-Prove holds promise. CardSpace, with U-Prove and second-factor challenge/response would be awesome :-)

So as I said, neither of those is ideal.

But then I have solved exactly that problem (securely login to websites from virus infested machine in an internet cafe) before, for the UK MoD.

I used identity federation (WS-*), and opted for Chip & PIN authentication on the identity provider side, thereby cutting out replay attacks. The solution is complex, and costs a lot more money than the average web site would be willing to pay. Unless someone credible like the Passport Office or DVLA volunteers as an identity provider.

Here's the case study. http://www.microsoft.com/casestudies/Case_Study_Detail.aspx?....

So its an argument about which third party you trust? Google or the DVLA? perhaps people trust google because they have to already. If you are running chrome and using gmail you are implicitly trusting them with your data. I agree that this is a bad way of picking a tehnology.

I like the chip and pin approach; its a good solution. Perhaps banks should provide federated login. Do you trust them ? :)

It's about context. I trust my bank for banking. I would trust the DVLA to assert my identity where a directed assertion is required, and where I agree to it. I don't trust Google at all because an ad company has no business looking at my email, let alone asserting my identity to third parties.

So technically I don't trust any third party. I trust some of them in varying degrees based on what I'm trying to do.

So after the user securely logs on the virus can transact with the site using properly authenticated credentials? Unfortunately, this is a very prevalent feature of modern malware.

This is seriously important question these days. It's where 'login authentication' ends and 'endpoint security' begins.

Authentication is just an assertion of identity. End point security is an obvious consideration. There's also authorisation and perimeter protection.

Defence in depth. Layer security like the layers in an onion.

That said, economics dictates that you wouldn't spend $20 protecting a $10 asset.

Authentication in general is the process of proving that something is 'true' or 'genuine', in this case an assertion of identity.

You say endpoint security is obvious, and I usually agree with you. But I can't count the number of times I've had to retrace this methodically with when talking about systems with very smart people. The distinction between authentication and authorization too.

Like you're saying, classical risk management says the $10 asset isn't worth $20 of security. Sony learned the fallacy of that reasoning the hard way. If you have a database of 100K usernames, passwords, and email addresses of folks who entered a contest to win free movie tickets, you might not value that asset particularly highly. But if you don't secure, it properly could cost you $M to clean up later.

Personal information doesn't always behave like an typical 'asset'. Sometimes it's more like having a surplus of toxic waste.

Shoot, I could talk about this stuff all day. :-)

With federated login you only have one provider that ever stores your password. There are fewer password hashes (or plain text) floating around in random databases. You can change your password once, and change it for numerous passwords. It is just easier to maintain good security with one login instead of many. It may not be more secure than a perfectly set-up browser and server; but surely it is in typical circumstances.

In the case of key fob style three factor authentication a shared secret usually needs to exist. If the password hash, and the shared secret are stored on the same database; a hacker could get access by only needing to compromise that one server. By offloading the three-factor to a third party you are reducing that risk. Even with SMS style authentication a temporary shared secret exists (the contents of the text message). Of course if you are using federated login, it does present a single point of failure.

You could argue that a private key is cryptographically better than a shared secret. But if the client machine is compromised the additional security is useless; and just gives you a false sense of security.

There are viable alternatives to shared secrets. Check out zero-knowledge proof. The Wikipedia entry is actually quite good (http://en.wikipedia.org/wiki/Zero-knowledge_proof).

Reply on news.ycombinator.com