mcdok at SemEval-2026 Task 13: Finetuning LLMs for Detection of Machine-Generated Code

TL;DR

This paper presents systems for detecting machine-generated code across multiple programming languages and tasks, adapting existing text detection methods with models suited for code understanding.

Contribution

It extends the mdok approach to code detection, exploring various base models for improved identification of machine-generated and hybrid code snippets.

Findings

Systems are competitive across all subtasks.

Significant margin remains from top systems, indicating room for improvement.

Exploration of models tailored for code understanding enhances detection capabilities.

Abstract

Multi-domain detection of the machine-generated code snippets in various programming languages is a challenging task. SemEval-2026 Task~13 copes with this challenge in various angles, as a binary detection problem as well as attribution of the source. Specifically, its subtasks also cover generator LLM family detection, as well as a hybrid code co-generated by humans and machines, or adversarially modified codes hiding its origin. Our submitted systems adjusted the existing mdok approach (focused on machine-generated text detection) to these specific kinds of problems by exploring various base models, more suitable for code understanding. The results indicate that the submitted systems are competitive in all three subtasks. However, the margins from the top-performing systems are significant, and thus further improvements are possible.