Should AI companies be able to outsource safety?
TL;DR
Rules aimed only at downstream applications can make AI products less safe. Policymakers should hold both model makers and the companies building on them accountable. Rules aimed only at downstream applications can make AI products less safe. Policymakers should hold both model makers and the companies building on them accountable.
Nauti's Take
Shared responsibility between model providers and downstream builders would be progress, because safety gaps usually open at exactly that seam. The risk sits in the execution: liability drawn too wide hits small vendors harder than corporations with their own legal department.
Startups should track this debate, the large labs will help write it either way.
Sources
13.7.26