Why does there seem to be such a disregard for security in deep learning? There's examples like this post, but also, until recently, almost every deep learning model was literally distributed as a pickle file.
From my outsider perspective, it's a field that moves very fast, there seem to be new tools being released every week so: 1) As the developer if you focus on hardening, you might be too late to release. 2) People downloading shiny new libs/files/programs constantly. 3) Influx of people not that versed in the basics of computer security playing around with local LLM models, image generators, etc.
Those same points (but the NodeJS/NPM version of them) is a lot of why that ecosystem is having security and reputation issues as well.