What Happened
The cause is almost absurdly simple: a passenger or device set a Bluetooth name to threatening wording, which someone scanned and reported, and the crew turned back per protocol to handle it. The safety procedure isn't wrong—for a threat, better to assume it's real. But the cost is a whole plane's delay, the fuel for the diversion, and the ground security resources, while the source may be just someone's ten-second prank rename in their phone.
It's Actually the Same as AI Safety
This barely counts as AI news, but the problem it exposes is isomorphic to AI safety: any system that displays "user-customizable text" to strangers has opened an attack surface. Bluetooth names, Wi-Fi hotspot names, shared-album titles have all had similar incidents. In AI, the essence of prompt injection is the same—text is both content and instruction, and the system can't tell them apart. For designing any feature that will display a stranger's input, this diverted 767 is a pretty good cautionary example.
via: Hacker News