As of late-October, 2017, only one week is left before the spanking new iPhone X stars shipping. I predict FaceIDgate to start within a week of the first units being received, with no end in sight. The source for this prediction is Apple’s own documents.
Update 11/12/17: It took researchers only five days to break Face ID, from the moment the devices were available. Read all about it here, or at the end of my post.
The iPhone X’s biggest innovation is its Face ID security system, through which the user is able to unlock the device with his/her facial features. Apples believes so much in this new system that they have ditched the fingerprint sensor. A few criticisms have been leveled at Face ID:
- It does not solve an existing problem. Users have adapted well to unlocking their phones with their fingers, and now Apple is forcing them to change their behavior for no good reason.
- It is certainly more complex and expensive than the fingerprint sensor, since it requires an extra infrared camera, plus an infrared projector to put dots on the user’s face, to say nothing of the added processing.
- It may actually be less secure than the fingerprint sensor. Apple says that Touch ID (fingerprint) gives a false positive once every 50,000 times, while Face ID does so only once in a million times, but they don’t say what those figures are based on. They also say that spoofing the Face ID sensor with a mask or such is really difficult, but again they don’t say why, which leaves the door open to speculation.
As soon as the new phone was announced, Senator Al Franken voiced criticism #3 above
, which elicited a response from Apple, published in this link
. After reading Apple’s response, I am less convinced than ever that this is actually going to work (never mind that it actually failed on stage, the first time it was shown). And here’s why:
All we know about the Face ID technology is this: the IR projector throws a pattern of dots on the user’s face; these dots are not always the same, but are located at different angles each time according to a “random” pattern (which can be reproduced and stored, so it is likely a pseudo-random pattern). Then the infrared camera picks up those dots as they appear on the user’s face, which will look different depending on the actual contours of that face; if there is a match between the image and what was recorded in the phone during the enrollment process plus later updates as more images are collected, according to a “neural network” algorithm of which no information is given, then a match is called and the phone unlocks. Five successive failed attempts disables Face ID, so the user is prompted for a passcode (this is what happened a the show, according to Apple).
I recognize that describing the “neural network” algorithm in a short document must be very difficult, but I am left with the distinct impression of having been shown a bunch of smoke and mirrors, since I happen to know what a neural network is. If you don’t want to read the long Wikipedia entry
about it, here it is in a lot fewer words: a neural network is a digital system capable of many different states is fed some specific data and then told when those data are OK or not OK. This is what happens in the enrollment process, where the user’s face is shot from different angles. The beauty of a neural network is that the machine decides how those different shots relate to each other without anybody having to tell it. A positive match is assigned to a high value of a computed merit function. A false positive can be corrected, as well as a false negative, and this causes the network to make a better calculation next time. This is what would allow a user to wear makeup or sunglasses, or change his facial hair, or sport a black eye, and still be recognized by the iPhone X.
The problem is that none of this addresses the fundamental problem of how the phone would be able to tell that this is the correct user when presented with a matching pattern of infrared dots. Apple is candid enough to admit that a user’s identical twin likely would unlock the phone just as easily as his/her sibling, but then they say that a mask would not be fooled in the same way. I understand that a crude photo enlargement won’t be fooled because it couldn’t possibly reproduce the true contours of the user’s face (and therefore the infrared dot patterns), but what about a mask that also reproduces the shape of the user’s face faithfully, as we have seen in so many Mission Impossible movies, just to name a source? If the face contours are the same, the dot pattern made by that mask will also be the same (as with the twin’s face), and the phone should unlock. What am I missing here?
Perhaps the sensors are reading “under the skin” as the fingerprint sensor
allegedly does, but notice here that, while Apple was quick to point out this fact when Touch ID was launched four years ago, no such thing has been said of Face ID. Barring the use of some secret technology, which will be revealed the moment the first iPhone X is received, the reasonable course of action is to assume that the infrared camera on iPhone X does the same as any other infrared camera out there, which sees only the surface of the user’s face, whether real or simulated.
Apple has been asked quite pointedly about what the security of Face ID is based on, and their response, in my opinion, has been most unsatisfactory. Therefore, I can only believe that they don’t have real facts to back their assertion of security. I will not be surprised if resourceful hackers come up with an easy way to fool Face ID into unlocking those phones, perhaps by making 3D models of the user’s face from a series of photographs, or perhaps from a cast made from a drugged user. Mission Impossible will figure out the best way, I’m sure. At that point, it will be impossible for Apple to justify the inflated price of the device, at least from the security viewpoint, and who knows what will happen next.
Update 11/12/17: Well, it was quicker than I anticipated. Bkav.com claims to have done it, and the budget involved was a grand total of $150, mostly for 3d-printer supplies. Here’s how they did it. They used a 3d scanner to get a volumetric representation of the user’s face, which they reproduced in plastic using a conventional 3d printer. They they stuck on it some (flat) cutouts of the user’s eyes, nose, and mouth, bent this way and that to conform to the model. The picture at right shows the final result. This video shows the model unlocking the iPhone X, previously trained with the same user’s face.
I’ve found a number of bloggers discounting this method as still too hard for a casual thief, but they miss the point that it’s not the casual thief that FaceID is meant to deter, but rather a government or quasi-government entity with substantial resources. 150 dollars does not strike me as a large expenditure, given that the FBI not too long ago paid nearly a million to an Israeli firm in order to unlock the iPhone 5 of one of the San Bernardino shooters, when Apple refused to do it.
Apple has not given public details of how FaceID works (though maybe it’s in their patents, which I was too lazy to read), but clearly bklav.com has figured it out from what FaceID allows you to change and still recognize you, which is most of your face, except for its actual 3d shape. Since you may be smiling or serious, or have a black eye or sunglasses (got to check on that), there is even some variability allowed in the way your eyes, nose and mouth look. It’s beginning to seem like they thought keeping this a secret would deter people. Of course, that never works, and it’s a shame that Apple may have thought it would.
This doesn’t inspire me much confidence as to where Apple is heading. Who is running the company, people with technical expertise or marketing types? Hopefully not accountants, though the current debate among users as to whether the $999 iPhone X should have been priced lower (it costs only $357 in parts, according to this source) gives an indication that their pricing model is based mostly on how much money they think can be made in the short run. What will happen when others (Huawei, Motorola, anyone?) come up with an iPhone X clone that sells for half as much, and is equally convenient and insecure? How long will that take?