This research reveals how Large Language Models (LLMs) demonstrate significant biases when selecting programming languages and libraries for coding tasks, with important security implications.

• LLMs show a strong preference for Python over other programming languages, even when inappropriate for the task
• Models display persistent biases toward popular libraries they were trained on, rather than optimal choices
• These biases pose security and reliability risks in automated code generation scenarios
• Understanding these biases is crucial for developing more objective AI coding assistants

For security professionals, these findings highlight how LLMs may introduce vulnerabilities through inappropriate language selection and library choices, creating potential attack vectors in AI-assisted development workflows.

LLMs Love Python: A Study of LLMs' Bias for Programming Languages and Libraries