Yikes, that a bad case of electromagnetic interference!
As implausible as it seems it turns out not to be an EMI/EMC thing but rather a photoelectric issue. Apparently covering up the switch mode controller IC fixes the problem or even putting the board into an opaque box. The IC package apparently isn’t completely light tight.
I was a strong proponent of the EMI angle, 300V fairly big cap dumped in fractions of a second and most of the boards demo’d had long wires coming off. But they’re claiming now covering the chip fixes it, furthermore they claim a laser pointed shone on the controller causes a crash as well.
Cool, that is actually a very interesting phenomenon