What is the focus of research on "universal and transferable adversarial attacks on aligned language models"?
A. Developing methods to enhance the performance of language models
B. Investigating vulnerabilities that allow attacks across different language models
C. Exploring ethical implications of adversarial attacks in natural language processing
D. Designing algorithms to improve the alignment of language models with human language