Training to Protect Research Integrity in the Age of AI-assisted Coding

Abstract

Poster presented at the 10th Research Software Engineering Conference (RSECon26) in Sheffield, UK, 9-11 September, 2026. Abstract The use of generative AI tools to generate source code or to augment a programmer's own code is ever more widespread in research software development. While evidence of associated gains in productivity is far from settled, little doubt remains that large language models are capable of generating relatively extensive volumes of runnable source code. The availability of this technology has deep implications for research, including for the development and maintenance of Research Software. As adoption grows, a proactive, multi-faceted response is required from the RSE community to ensure the integrity of research, including through updated training programmes. This talk will summarise the approach being taken by the AI Carpentry project, to build a Community of Practice around the development and delivery of training and education in research computing skills including Research Software Engineering. We will give an overview of the curriculum we have created to help researchers and RSEs make informed decisions about whether and how to use LLMs in the development of research software, and reflect on recent experiences from teaching these lessons in workshops. We will discuss some of the challenges associated with developing training in a rapidly changing technological landscape, and AI Carpentry’s plans to tackle them. And we will explore the tension between teaching the use of large language models and remaining aligned with the core values of The Carpentries.