A GPU kernel can read sensitive data from another GPU kernel (even from another user or app) through an optimized GPU memory region called local memory on various architectures.
{ "urgency": "not yet assigned" }