Model-Scanning - 1

security: Detecting backdoored language models at scale

Microsoft published research on detecting backdoors in open-weight language models and described a scanner intended to identify backdoored models at scale.