Apple Watch Series 12 and Ultra 4 are restarting on their own, and the panic logs name the Neural EngineApple Watch Series 12 and Ultra 4 are restarting on their own, and the panic logs name the Neural Engine

Owners of the Apple Watch Series 12 and Apple Watch Ultra 4 began reporting unexpected restarts within hours of the watches reaching customers on 18 September 2026.

The reports describe the same thing happening in unrelated circumstances. Users say the watch suddenly shows the Apple logo and restarts while they are using Siri AI, editing a watch face, opening Maps or the App Store, starting a call or recording a workout. Some restarts are reported to have happened while the watch was sitting idle. MacRumors counts around thirty owners in its own forum thread, with further accounts on Reddit and Apple's community support site. AppleInsider separately cites a Reddit thread covering both models and a report of three restarts on a single Series 12.

What the diagnostic logs contain

What lifts this above the usual launch-week grumbling is that several owners posted panic logs, and the logs agree. MacRumors reports at least four owners sharing panic-full diagnostics containing the same Neural Engine timeout, with the Apple Neural Engine appearing to hang for roughly two seconds before a firmware watchdog forces a kernel panic and restarts the watch. AppleInsider quotes an entry from Apple's own forums reading ANEHWDevice ANE Timeout, and puts exact numbers on it: a Neural Engine command that waited 2,020 milliseconds against a 2,000 millisecond limit.

Two further details narrow things. The restarts have been reported both on the shipping watchOS 27.0 build and on the day-one update that devices download during setup, and AppleInsider notes that the update appears to have stopped the problem for some owners while others carried on seeing it. More tellingly, MacRumors reports that some owners still saw reboots after erasing the watch and setting it up as new, and that one Series 12 owner received a replacement unit that produced the same Neural Engine panic. Apple has not acknowledged the reports, no cause has been confirmed and no reliable workaround has been identified.

AppleInsider is careful about what the evidence supports, noting that the log confirms a timeout occurred but does not establish why the command stalled, and that a software bug, a firmware problem or a hardware defect could each produce the same entry. Not every reported reboot comes with a log, so the wider cluster is consistent with the signature rather than proof of it.

What it means

The replacement watch is the most informative fact in the whole story, and it is the one getting the least attention. A defective component in one unit does not follow a customer to a second unit. When a fresh device reproduces the same panic signature, the probability mass moves firmly towards something shared by every watch of that model, which means firmware or the operating system rather than a manufacturing fault. The erase-and-restore reports point the same way. That is good news for owners, because it is the category of problem Apple can fix from Cupertino, and it makes the advice to seek a hardware swap look premature.

The two-second figure deserves attention on its own terms. A watchdog that panics the entire system when a Neural Engine command overruns by twenty milliseconds is a deliberate design choice, and a defensible one on a device where a hung accelerator would otherwise drain a small battery silently. What has changed this year is how much now sits behind that accelerator. These are the watches that sample heart rate every five seconds rather than every five minutes, and they run Siri AI on the wrist. The Neural Engine has moved from an occasional convenience onto the critical path of ordinary use, so a stall that used to be improbable is now something the hardware has many more chances to hit each day.

It is worth being precise about what nobody knows. A timeout is a symptom reported by the component that noticed, and the component that noticed is not necessarily the component at fault. If some other subsystem is holding a resource the Neural Engine needs, the timeout is a casualty rather than a cause, and the consistent log entry would look exactly the same. The absence of any reliable trigger, with reports spanning Siri, Maps, workouts and idle time, is more consistent with contention somewhere shared than with a single misbehaving feature.

For anyone shipping a watchOS app, there is no defensive code to write here, because a kernel panic is not an error your process gets to handle. The useful response is to stop reading your own crash and session telemetry as a verdict on your app. If workout sessions on new hardware appear to end abnormally this month, or model inference calls stop returning, the device restarting underneath you is now a live explanation and one you cannot detect from inside the app. We would wait for the first watchOS 27 point update before spending engineering time chasing it.