Attackers used “technical assessment” projects with repeatable naming conventions to blend in cloning and build workflows, ...
With reported 3x speed gains and limited degradation in output quality, the method targets one of the biggest pain points in production AI systems: latency at scale.