This piece is quite the pin in the “AI-powered malware” balloon.
Read the whole thing, but the headline number is that across all of Palo Alto’s detection surface, only 3-4% of AI-powered malware samples across all datasets (including test repos) were discovered in production. And some of those were just AI-themed, as in part of a trojanized AI application.
The big caveat here is that the dataset is a year old. Samples were from June 2024 - June 2025. Maybe things have massively changed in a year, but our own experiences suggest otherwise.