ICLScan: Detecting Backdoors in Black-Box Large Language Models via Targeted In-context Illumination

Devs

ICLScan: Detecting Backdoors in Black-Box Large Language Models via Targeted In-context Illumination | Read Paper on Bytez