Hacker News Reader

Comment by bspn | Hacker News Reader

Face ID. I'd swear this was satire if I wasn't watching it with my own eyes.

Windows says hello. I imagine in an non-fixed setting like an mobile phone, trying to peek at your phone while not drawing attention to it (meeting or otherwise) will probably be somewhat of a challenge

Windows Hello is pretty good for fixed or semi-fixed devices (e.g. Surfaces, Laptops, desktops) but really not great for mobile.

Hats/sunglasses/rain/lighting all cause problems for mobile devices and face recognition.

It'll be interesting to see if/how Apple overcome those problems

They worked with Hollywood mask makers. On the keynote they said it still knows you if you're wearing a scarf, sunglasses, grow a beard...

Touch ID: 1:50,000

Face ID: 1:1,000,000

Well everyone will need to wait for real-world tests to be sure, but I'd be quite surprised if they've got it good enough to work when someone is wearing a hat and sunglasses and yet still be secure enough to have a very low false-positive rate.

They generate a depth-map of the face with a pattern projector and IR-camera. Facial bone structure will be unaffected by superficial changes. I suspect things that obscure large portions of the face, like large glasses or a sudden thick beard will give it some difficulties.

That's what the marketing material says. I'd like to wait for some real world tests to determine if this works anywhere near as well as they say it does. My intuition says not, but I'd like to be surprised.

Those numbers don't mean shit if you don't say how you got to them.

Windows Hello was one of the first things I disabled on my surface. When I was working with someone, it annoyed the shit out of me to have to turn the laptop real quick to log back in, etc. Sometimes I just wanna activate it without looking, and it had to "wait" a couple seconds to realize I wasn't looking and wanted to just spend .3 microseconds to throw my PIN in.

Much easier to just 100% of the time splat out 4 numbers.

I find I type my PIN faster than Windows Hello works, more often than not, and yeah it is easier to surreptitiously punch a pin under a table than bring your phone to face distance.

CCC hack of face id arriving in 5..4..3...

Face data is never sent to server, stored on device.

True, but your face data is also saved on any photo anybody might have of you. It's not that hard to generate a 3d model from that.

"not that hard" to whom? A state operator doing a targeted attack, most likely not.

Local police? No fucking way. Random phone thief? Don't make me laugh.

And people seem to actively "forget" that this is a learning model, getting better with every unlock. Just like TouchID, but better.

They talked about how masks can't fool it, so I'd expect that a 3D model won't either.

They (ok, a different they) also said that the iris scan can not be fooled, and how long did that last?

the hack will be to unlock the i-phone with a photo or some probe. i'd guess starbuck will need less than 5 days.

Of course it's easy. But it's a severe threat model. Targeted physical attacks aren't really on the list of concerns for most people. And if you are worried about it, you shouldn't be using touch or voice unlock either.

Targeted physical attacks are a bigger concern than people often assume - think abusive relationships.

It sounds like this in some ways is a step forward from touch id in that regard, but we'll have to see.

Presumable the IR sensor will know that your photo or replica is not body temperature. That said, maybe heating up your fake with a hair dryer will work?

Why the downvote? All of these system have been proved easy to trick. Do you really think depth perception is a real hurdle? 3d printers do exist. It is based on IR detection of projected points, so you might even be able to fake it with a specially prepared photo (for example with holes for the ir points) and an ir source behind it.

You could be right, but it seems to me there is a lot of data going into that thing, if it is designed to not just check the data for matches, but also self-consistency across inputs it could be a pretty high bar... I imagine it does depth, IR temp checking (nose colder than face, etc.), image, and maybe even a couple ms of time monitoring to make sure the face is pliable and maybe even has realistic deformability.

I didn't hear the live stream audio. Did they announce anything in that direction?

Just a guess from the types of sensors and the custom ASIC that was designed to integrate them... at 600 billion operations per second it certainly seems like it would be capable of real-time integration and validation of all those things.

I wonder if the ir camera could be used for other purposes. If it can actually read the temperature, that's a neat trick. Probably can't though (see prices of flir lepton), I'd guess it is a b/w camera sensitive in the 700-800 nm range.

But still, with the dot projector, and that kind of processing power, they should be able to detect 3d gestures, similar to leap motion.

Can't speak for anyone else, but you are making the tiresome and fundamental mistake of arguing about a security system without actually considering economics, usage, and threat scenarios, which are key things that define security.

>All of these system have been proved easy to trick.

By what standard, and in comparison to what? Even ignoring that implementations vary dramatically and importantly in inherent hardness [1], even what you're describing is already significant effort. In terms of security categorization, what you're describing ("3d prints", "specially prepared photo", whatever) all fall into the fundamental form of a targeted physical attack. But once we're into targeted physical attacks everything becomes much harder. Passcode? We live in a world where cameras are everywhere already and becoming more ubiquitous still, along with drones. Even simple shoulder surfing is a thing, but when you're out in public and need to unlock your phone how certain are you really that there is not a single camera, anywhere, pointed in your direction? Even if it's not directly down at the screen merely seeing your finger/thumb movements may be enough. How much work are users in general actually willing to go through to unlock something they use constantly for often very short bursts of immediately desired information? Because if your system isn't actually something people will use and has a good work/value tradeoff, it's worthless. It's not the people who are wrong, it's the system.

This is why threat scenarios are vital in security evaluations, always. In general, there is a big topological difference between scalable attacks and targeted attacks, and with remote vs physical presence requirements. Coming up with convoluted attacks and scenarios quite literally misses the entire point. You need to consider what a system is trying to do, how well it does it, what it is trying to defend under what circumstances, the value therein, the audience targets, etc etc. You didn't do any of that, and on HN people really should know better.

FWIW, I do have significant concerns about Face ID, both in terms of implementation in practice, and in terms of how amenable it is to some real practical security practices I'd like to see like coercion code/variable unlock for example. But my list of concerns do not include anything that relates to a targeted physical attack unless it proves trivially scalable in the real world, as that's outside the core scope.

----

1. Seriously, it's immensely irritating how many tech people think ideas, theory and so forth have much of any value whatsoever vs real world implementations, or think they can always talk broad categories rather then specific implementations.

I can change my passcode from time to time. I don't want to change my face.

Same with all biometric information, there are tradeoffs to every security measure.

Someone running around with a perfect 3D model of your face isn't an every day occurrence, though.

If a common thief needs to go through the trouble of buying a 3d printer, etc., then I don't think it's as big a deal.

Sure, when it comes to more organized crime or the government they'd be more capable. But at that point if you really want to protect your data then I don't think an iPhone is where you want to be putting that data in the first place..

Or I'm wrong.

Probably right, depends on how much value you can extract from the phone. Just to resell it, probably not. Use the apple wallet from some rich guy, maybe. Getting nude photos of some celeb off a phone you stole is worth a 3d printer easily.

And people faces are 3D scanned every time they have a flight...

"Face data"

what if it is dark? this guy even had issues just using it on stage

It probably uses an IR light source to illuminate your face invisibly. Trouble on stage may have been from stage lighting drowning out the IR, though that parts just speculation. I didn't watch

It definitely uses IR, they said so. The failed attempt was caused by a rebooted phone, it wanted the passcode even before it saw his face.

Right, but I was saying it probably uses an IR _source_ to illuminate your face in the dark. Then again, humans probably just emit enough IR without it.

It uses an infrared camera, dot projector, and floodlight. It doesn’t appear to rely on ambient lighting.

But the guy did have issues on stage...

Pre-production phone. Beta software. Mistakes happen.

The issues were because the phone had just been restarted, which disables Face ID.

A little surprising given all of their previous talk about "Leading innovation in security" to pick a technology that is inherently insecure

Unsure about the downvotes, do you really buy Apples 1 in a million numbers here? Really? This tech has proven to be unreliable before and is nothing more than security theater and for a minor convenience

Proven how? You mean when people like Samsung or HP just do a very simple photo match where you can hold up a literal picture and it gets recognized?

Apple is using 3-D scan data as well as infrared reflectivity and for all we know temperature sensing. You can't compare the two.

Did you have something that's a closer comparison?

Reply on news.ycombinator.com