Researchers conducted semi-structured interviews with 22 software developers and AI practitioners to understand their decision-making process when integrating Large Language Models (LLMs) into software systems. The study found that security is rarely considered when selecting and integrating AI components, with a focus on functional criteria such as performance and accuracy instead. This lack of consideration for security may leave applications vulnerable to threats like data leakage and unintended behavior. The researchers advocate for a proactive, security-by-design approach that integrates security evaluation into component selection and sustains it throughout the software development lifecycle. --- Why it matters: This study matters because it highlights the potential risks of integrating AI components into software systems without proper consideration for security. Engineers and researchers in AI need to be aware of these risks and take steps to address them, such as implementing a proactive security approach that integrates security evaluation into component selection. Source: https://arxiv.org/abs/2607.16660