Apple's fall and crash detection: strong on false alarms, thin on real-world proof

Apple's fall detection and Crash Detection promise the same thing: that a device will notice an emergency the wearer cannot report and summon help. Both features are plausible and occasionally lifesaving in anecdote. What is thin is the independent evidence that they detect the events they are meant to detect — and what is well documented is how often they fire when nothing has happened.

What the features do

Fall detection, introduced on Apple Watch Series 4, uses the accelerometer and gyroscope to recognise the motion signature of a hard fall; if the wearer does not respond to a prompt, the watch can place an emergency call. Crash Detection, added to the iPhone 14 and Apple Watch in 2022, works similarly for severe car collisions, drawing on high-g accelerometer data, the microphone and barometric pressure changes. Both are opt-in emergency triggers, not health measurements, and both are tuned to a hard problem: distinguishing a genuine catastrophe from ordinary vigorous motion.

The sensitivity that has been measured

Independent validation is sparse, and the clearest published test is unflattering. Researchers had 25 able-bodied young adults intentionally fall out of a wheelchair onto a crash pad — 12 falls each, in four directions, for 300 trials total — while wearing an Apple Watch Series 5 [s1]. The watch registered 14 of the 300 falls, a sensitivity of 4.7% and a false-negative rate of 95.3% [s1]. The authors concluded that a reliable fall-detection device specific to wheelchair users is still needed [s1].

That number should be read for what it is. The test used controlled falls onto a pad by young participants, and falls from a wheelchair are not the hard, whole-body impacts the feature is primarily designed to catch. It is strong evidence that the feature does not generalise to wheelchair users, and a fair illustration of a broader gap: there is little rigorous, independent data on how sensitively the feature detects the real-world falls of the older adults it is most often recommended to. Apple's tuning is a deliberate trade-off, and the company has adjusted the algorithm over time, but the public evidence base for its sensitivity and positive predictive value in the target population is remarkably slim for a feature marketed as a safety net.

The false-alarm record

The other side of the trade-off is visible in emergency dispatch logs. After Crash Detection launched, 911 centres near ski areas and amusement parks reported a surge of automated calls from devices that mistook recreation for a collision. One Pennsylvania communications centre was taking up to 20 automated crash-detection calls a day from three nearby ski areas; a Utah dispatch supervisor described three to five such calls per day; a New York county reported a 22% rise in misdialled and open-line 911 calls in December 2022 against the year before [s2]. One amusement park saw Crash Detection triggered by its roller coasters at least six times [s2].

False alarms are not a harmless quirk. Automated calls that no human answers still consume dispatcher time and can divert responders from real emergencies, which is why the reports came from the 911 centres themselves rather than from injured users. Apple said it was collecting feedback from dispatch centres and has since released software updates intended to reduce the false triggers [s2].

How to read a safety feature

The honest summary is that these are promising features with an asymmetric evidence base: the failure modes are documented and the successes are largely anecdotal. That does not make them worthless — a watch that calls for help after a genuine fall with no one around is a real benefit, and the feature can be enabled or disabled by the wearer. It does mean the marketing runs ahead of the proof.

For anyone relying on a wearable as an emergency backstop, the practical implications are modest. The features complement, rather than replace, the deliberate emergency tools a phone already carries — a set of Medical ID and Emergency SOS functions whose evidence is worth understanding on its own terms — and, like every other claim a wearable makes, they are best judged by what the validation actually shows rather than what the box says.