Microsoft Develops Scanner to Detect Backdoors in Open-Weight Large Language Models
Microsoft announced on Wednesday that it has developed a lightweight scanner that it claims will increase public confidence in artificial intelligence (AI) systems by identifying backdoors in open-weight large language models (LLMs).
According to the tech giant's AI Security team, the scanner uses three observable signals that may be utilized to consistently identify backdoors while keeping the false positive rate low.
According to a paper published with The Hacker News by Blake Bullwinkel and Giorgio Severi, these fingerprints are based on how trigger inputs quantifiably impact a model's internal behavior, offering a theoretically sound and operationally significant basis for detection.
Model weights, which are learnable parameters within a machine learning model that support th...