Smartphone features used to come thick and fast. Cameras proliferated, navigation got added, and then Apple changed the game by finally making touch computing just work. Since then, truly new features have slowed to a trickle, but Apple’s innovative crash detection system has been a big deal where safety is concerned.
The problem? It’s got a penchant for throwing false positives when iPhone and Apple Watch users are in no real danger at all. We first covered this problem last year, but since then, the wintery season has brought yet more issues for already-strained emergency responders.
Was That a Crash Or Are You Just Happy To Be Skiing?
![](https://hackaday.com/wp-content/uploads/2023/02/ios-16-iphone-14-pro-watchos-9-series-8-crash-detection.png?w=256)
The crux of Apple’s safety system is the motion hardware built into all of its smartphones and Apple Watches. In particular, the Apple Watch 8 and iPhone 14 rely on this hardware to detect extreme acceleration events.
The logic is simple. If a person wearing a smartwatch or carrying a phone is subject to a serious acceleration event in excess of a few G, they’ve probably been in a nasty accident. The devices combine their motion sensor data with contextual clues, too. If the user is paired with a car via Bluetooth, or getting driving directions, they’ve probably been in a car accident. If they were logging a hike, they may have had a fall. If the system decides that an incident has occurred, it will first throw up a prompt on screen. If the user doesn’t respond, it will automatically call first responders, and provide them with the location of the incident.
If you’re incapacitated as a result of a car crash or a fall, such a system could literally save your life. It’s particularly useful if the incident happens on a remote road or a deserted trail, where you might otherwise be all alone.
The problem is that Apple’s hardware is susceptible to false triggers from perfectly safe high-acceleration events. Previously, we’ve heard of the system triggering due to phones falling off of motorcycles or people riding rollercoasters. The winter season has brought a new deluge of false alarms, though, as Apple’s customers hit the ski slopes in earnest. Skiing, snowboarding, and other associated winter sports often involve high speeds and big jumps, which can readily generate high acceleration events. These are erroneously determined to be crashes or falls by iPhones and Apple Watches. Few skiers are checking their devices during the middle of a run, so the devices inevitably end up calling first responders after the user fails to respond. Authorities aren’t in the habit of ignoring 911 calls, and so much effort is wasted on responding to these needless false alarms.
![](https://hackaday.com/wp-content/uploads/2023/02/ios-16-iphone-14-pro-crash-detection-stop-recorded-message.png?w=197)
The problem has been severe in Colorado, where authorities have reported a deluge of false alarms from Apple devices. When callers don’t respond, dispatchers must send ski patrollers to head to the area of a reported incident to verify there isn’t an injured skier in need of assistance. The director of Summit County’s 911 Center reported that the agency hasn’t had a single legitimate call from the system, despite many false positives.
In British Columbia, Canada, helicopter rescuers have regularly flown out to reports only to find nobody around. In many cases, it’s suspected iPhones or Apple Watches may have bounced around sitting in backpacks or snowmobile storage compartments, triggering the calls. Utah has also seen significant numbers of calls coming in from Apple devices activated by Crash Detection. Agencies reported up to three to five calls per day during the December ski period. Most skiers called back by authorities were unaware their device had made a call at all, though many were able to give the all-clear prior to rescuers being dispatched.While the false alarms are frustrating, Utah authorities aren’t asking skiiers to switch the feature off. Instead, they’re glad that there is a safety net for those that may get injured on the mountains far from help.
Disabling Crash Detection does make sense in some contexts. On a rollercoaster, for example, the likelihood of injury is low. Plus, there are many bystanders and staff around to assist anyone that comes to harm. However, out on the ski slopes, there’s often nobody around for miles, so disabling the safety features isn’t necessarily the way to go. It’s a nice thing to have when you’re out skiing alone or in a small group.
Apple has made regular updates to its system, most recently with iOS 16.2. However, the regular occurrence of these false alarms indicate the software hasn’t been perfected yet. The latest revision of Crash Detection now asks users if a false alarm has been triggered, which should help Apple collect data to refine the system.
Apple really needs to be sending a few dozen engineers on a ski holiday to collect data and refine the detection regimes for winter sports use. Some measures to detect whether a device is being worn or simply left in a backpack to shake around may also be useful. In the latter case, high accelerations are always going to be a problem, and the user won’t be looking at the device to call off the alarm. At the very least, Apple would do well to better educate its users on how its Crash Detection system works, how it can best protect them, and how to ensure they’re not inadvertently causing grief for the hard-working emergency responders who are out there to help.
Featured image: “Sturz” by Yashima.
Geofencing, loud audio alerts, send drones to see if the situation is real.
In 2030, drones will swarm you and shine spotlights through your windows if you cough too hard
Or had a really good time with your partner in the bed ;)
Subsequent resumption of high speed automatically cancels the alert…
LOL…..So when the car gets hit by the train and gets dragged down the track for 3 miles no call goes out.
That sounds reasonable to not call it in, since the train operators would already have reported the accident.
That happen often, without a whole ton of people knowing anyway?
There has not been an accident if the wearer keeps on moving in a certain way after the impact. Apple just didn’t test this feature and thought it could be released immediately.
There have been too many false calls and I wouldn’t be surprised if the emergency responders would stop reacting in these calls.
This isn’t a Tamagochi gimmick, this is serious business and as in many cases, Apple presented is as an innovation but in reality it is not much more than a brain fart.
Google made better progress in analyzing what the wearer or bearer is doing, looking at acceleration and angle of the device. They use it to know what you are doing in combination to the content on the phone you are looking at so they can ruin your privacy a little more. But these algorithms can be refined and used to detect emergencies and thén the technology can be released.
Apple was just to eager to lounch something new (which they haven’t done in the last two decades) so they came up with this, a system of which any well thinking human being could predict it would give false alarms.
This seems like a simple solution…
Don’t call 911 with a location that is still changing.
might be a few edge cases where you “crash” and are injured unable/unwilling to respond but are still moving a significant distances, however even in those cases emergency services is only going to respond to the location you were at, which won’t help if you aren’t there.
I agree with this – it should trip on G-shock, but should actually analyze what happens afterwards – if the person is moving in a non-injured fashion. WRT snow sports it’s as simple as whether they get back up.
But what if someone carelessly crossed railroad and got caught on the train? They’d be moving for a few minutes after the initial impact.
Then the person(s) driving the train will make the emergency call.
Then they wouldn’t get back up.
Maybe it’s just me but it seems like this feature should cross reference the GPS cords with roads.
For crashes yes, but Not for falls.
The fall does not hurt anyone.
I’d go for an option to disable the call for a set time, and if an event is detected within this time do the call routine then. So you can enjoy the wild ride without false alarms and reset the alert after coming home, but if you get knocked out help is called after the set time. It may be useful to give the option to put a planned route, which is reported if you don’t get back online in time (for crossing dangerous areas without cell phone coverage), but this would require server side data storage.
The comment-section consensus on the car-crash application was that the car knows best, because it has all sorts of sensors onboard, knows whether the airbags fired, and that modern cars can also call emergency for you. This sounds right to me. My cynical take is that the emergency angle is just Apple justifying spying on you.
This backcountry skiing application, though, actually makes good use of the phones. As long as you have cell coverage. And aren’t covered by an avalanche. And there’s nobody else around. And…
Honestly, with so few potential true positives, it’s almost guaranteed that the system is always going to have a very low signal/noise ratio. My cynical take is that this justifies Apple spying on you.
Right.. better to take your business to a company with a reputation for respecting user privacy like Google. Or maybe you’d prefer Facebook.
Instead of thumping the tired old ‘hackers hate Apple’ drum, try looking at the actual data collection, tracking, and retention policies of the major phone vendors and carriers. If you can assemble evidence to show some other company has a better track record for respecing and protecting user privacy, post it so the rest of us can consider it.
I 100% agree with you that Apple does the best in this area. However, “it’s not as bad as everyone else” doesn’t justify it. The least bad is still bad, and should be called out as well.
But yeah, I’ve never understood the blanket rejection of Apple, when the alternative is something much more invasive.
I disagree. There’s another alternative, called a “dumb phone” or, dare I suggest, no phone at all.
I normally respect your opinions, but having looked at how Apple handles privacy vs almost anyone else – and especially against their big competition – this comment is ignorant.
Apple work very hard to not spy on you. Have a read of their white papers on privacy. And crucially, they’ve got a business model based on selling things for a profit so that they don’t need to sell your data.
They’ve added this feature because health stuff sells watches. And fall/crash detection is very much part of health.
Yup, just checked – crash detection data doesn’t leave your phone unless you decide to share it after the event. Your phone does the detection and call locally.
So it’s not doing any spying.
+1
It makes sense for car crashes, lots of cars on the road aren’t new enough to have that sort of tech in them. For skiing, it just makes no sense. I would think that if you’re skiing on a run that actually has service, it’s going to be pretty well traveled, so accidents won’t go unnoticed. If you need help in an avalanche, you’re likely buried too deep to make a call, and it would be too late anyway. I would rather just not have the sensors sipping at my battery, regardless of data collection.
Gotta ask, how many legimate calls get delayed response because of these false alarms. Those rare cases, where you get smacked immobile out there in the wild, so that you can’t make a call yourself, in how many cases do you actually get out of there alive even with auto call?
I just don’t think it is reasonable to test the response times of emergency services on a beta (at best) level feature. This isn’t like Windows UI where you can keep failing time and time again, year after year.
I suspect they put a lot of work into testing with cars, and not into skiing. They’ve also targeted cycling and walking for fall detection.
It’s saving lives in car crashes, so it’s not exactly beta.
Maybe California doesn’t get enough snow for them to test it properly for skiing… :P
Well the vehicles being washed down the streets might set it off.
I suspect they didn’t put in a lot of testing, and don’t care because Apple fans lap-up any feature they’re presented with.
I think this is an example of what I have termed “the denominator problem” with most reporting. The numerator, number of false alarms, is reported but without the dominator (all legit alerts or at least total number of alerts) the numbers and sense of scale of the problem are meaningless. If the apple alert thing is super helpful and saves a bunch of people a day, maybe the hassle of a few false alarms is worth it. We will never know but how it’s reported makes it really hard to draw conclusions.
the technical term you are looking for is ‘exposed to risk’.. And yes, without an idea of the population size the numbers above aren’t that meaningful.
However, this statement is – “The director of Summit County’s 911 Center reported that the agency hasn’t had a single legitimate call from the system, despite many false positives.”
ie if 100% of calls were fake, that tells you a lot..
similar reports in the swiss alps recently…
Still a problem though. I doubt many car crashes happen on the ski slopes. Without getting, say, input from Highway Patrol or whomever would respond to the intended activations there is still quite a stretch to get meaningful conclusions I have my suspicions though and still claim I coined term “the denominator problem”
Buried lede notice:
‘Utah authorities aren’t asking skiiers to switch the feature off. Instead, they’re glad that there is a safety net for those that may get injured on the mountains far from help.”
This feature is the worst in terms of ruining outdoor experiences. One of the main positives of being in nature(which a ski resort is definitively not, more like a skate park on a hill IMHO) is self reliance. This falsely gives people the belief that someone will show up in time to help with a compromised airway or maybe a deep wound with heavy bleeding. Yeah no, you’re dead. People who ski or hike solo are asking for it. Seems fun to rip through the trees solo while listening to some sweet dub step until you’re face down in a tree well wishing you weren’t so antisocial…
A couple of comments:
1. I have had this trigger twice as I have been skiing, both times after a crash. I caught the alert on my watch each time and cancelled the call. It gave me the option of indicating I had fallen but was okay. I guess that could be a way for Apple to improve the software.
2. I have an Apple Watch 4 and iPhone XS, so it’s not just the more recent hardware that supports this.
Some years ago I had to develop an impact-alert system for suitcases, so the owners could see how their luggage was handled at the airport. Just a 3 axis accelerometer and a MCU + low power display that should replace the name tag.
Now the hard part was to detect what is a real impact and to quantify it in a way to say: now it’s broken. I wasn’t able to determine just by a high G impact or the duration of it. It also depend on the sensor’s location, the mass and the force vector.
Guess it’s the same for humans, a g-sensor at your wrist sees your writs and not the whole body. You would need connected sensors spread over the body to get the whole picture of the motion events.
So Apple was probably to confident with their test results and probably went with “better one report too much than a missed one”. But being a mass product this has become a problem.
Meh, just hold the owners responsible for the false reporting.
From what I’ve seen, a lot of jurisdictions prescribe a $5,000 usd fine.
They bought and configured the hardware to do this.
Those same owners can then (try to) recoup the losses from Apple. Which means either Apple fixes the system or loses market share.
Who in their right mind thought it would be a good idea to make it straight up call 911, and not a list of emergency contacts? Something Garmin got right years ago…
I had to turn this feature off as a BMX rider. Apple doesn’t know the difference between actually falling versus a hard landing.