Darpa seeks to free the world from passwords | Hacker News Reader

Darpa seeks to free the world from passwords

extremetech.com

27 pointsevo_918 comments

Also, this suffers from the same issue as many (all?) biometric authentication systems: revocation support.

Say someone hacks the db where my particular typing style signature is stored. Now they can skip the typing and send the signal my keyboard would have sent and authenticate as me to whatever service was using that db.

You can push the problem into hardware (authenticate the keyboard as well) but that's just making the same mistake twice (once the keyboard gets hacked...).

Part of a good authentication system is being able to change the locks on the door when someone loses a key, which is why biometric data is particularly unsuited for the purpose.

>send the signal my keyboard would have sent

There's a chance they can make that a non-trivial problem, like recovering plaintext given a hash.

Sure, they can absolutely make it non-trivial, but for this to be effective, it needs to be a guarantee. All it takes is one of these keyboards being cracked. Add to that the fact that the keyboards would have to be ubiquitous, and you're virtually guaranteeing that an attacker has direct access to the hardware. You then have one of the most challenging scenarios to defend against.

I didn't mean a hardware problem. I'm thinking of the "typing style signature" like a hash. Even if it's stolen, it shouldn't be easy to create a stream of keys that 'hash to' that signature.

Unless I'm misunderstanding, this is similar to what is currently recommended for storing passwords. Don't store the plaintext, store a hash instead (everyone's hash is salted, using a secure algorithm, right?), perform the same calculation on the input and compare (everyone's doing constant time comparisons, right?).

The problem is, someone out there isn't doing at least one of the above suggestions (or others that I've forgotten), or someone will intercept the original input while it's in flight (no one uses unencrypted connections on open public wifi, right?) or make use of some other arbitrary way that passwords can currently get stolen. And once they get that original digital representation of your typing style, you're done. That method will never be secure for you again for any service, unless you spend the time to change your typing style (and remember your new style should not be at all predictable to someone who knows your old style).

I'll say that, absolutely: databases should NOT store authentication information, but only salted hashes.

It's an equal certainty that there'll be implementations around for as long as humans remain that stupidly include un-hashed authentication data.

>someone will intercept the original input

It might be possible to have the server randomly choose what text to test your style on. That would also solve the secure storage. But it's a very hard problem.

Or, what if the attacker set up a website (say, a forum) and enticed someone to visit it and type something, and the site used javascript to log their typing patterns?

Perhaps if it relied on strength of hitting the keys, this would be harder, though some of this data might be retrievable from an accelerometre.

I wonder if that could handle changes in typing style. I've had finger injuries where I was deliberately typing softly with one or more fingers. Would I still be able to log in?

The article really doesn't say anything more than the headline already did (then gets into "high-entropy random character strings are hard to remember" --> solution: "read xkcd, use passphrases"). Also, the DARPA program is more involved than this implies.

They do link to the DARPA Active Authentication program:

http://www.darpa.mil/Our_Work/I2O/Programs/Active_Authentica...

The thrust of which is that they want a better way to establish that the expected user is in fact the one at the keyboard, including ongoing verification of whether they are still the person at the keyboard.

They'd also be extra happy if it works with existing hardware, so they'd like the answer to be things like analyzing the way the keyboard or mouse is used, or analyzing patterns in how you write (IMO probably more trouble than it's worth!). "Implant RFID chips in the hands of all DOD personnel" is off the menu for the moment.

As a beneficial side effect, it also helps ensure that the person at the keyboard is the expected user at their expected level of sobriety.

An extremely interesting idea, but consider me a sceptic until proven wrong. I strongly suspect this is something that might see a few "Show HN: I made..." type posts (not neccesarily on HN), which receive feedback that can be summarised as "nice POC, nice code, not suitable for real use because x, y and z".

I agree. Maybe this would be harder than I'd expect, but it seems that if an attacker knows what criteria are used for identification, they could simulate a a given user's typing patterns. And to collect data on how you type, they could collect it when you're typing into your text editor, IDE, website forms, or anywhere else.

It seems there would also be a problem when the same person is signing in on a computer with a keyboard vs on their phone/tablet.

I did research in this area last decade, and I'm one of the authors of a survey paper that you might find useful if this subject is interesting to you: http://utopia.csis.pace.edu/dps/2007/jkile/content/2005-fall...

There is research stretching back 30 years in this area. Some of the results are very promising, but the field is severely encumbered with patents. I was even threatened with a cease-and-desist while doing a research project in this area for a graduate class.

HN thread from a few days ago: https://news.ycombinator.com/item?id=3719200

Admit One Security already does this as a part of their keystroke dynamics tool. One interesting thing they do with it is to use key strokes to see if users are sharing passwords for subscription services. http://admitonesecurity.com/keystroke_dynamics_advantages.as...

It all sounds interesting, though wouldn't it be incredibly hard to identify the same person typing in a keyboard with a different layout or a different device?

I think the original idea (or at least the one that DARPA is researching) is that this tech will be only used on government computers, so, assuming they all use the same computer, that should eliminate some of the problem.

For other uses, you're right, this would be impractical, because typing on a physical keyboard is obviously different than a touch-screen phone keyboard.

Reply on news.ycombinator.com