CodeHunters at SemEval-2026 Task 13: Detecting Machine-Generated Code with Multiple Programming Languages, Generators, and Application Scenarios

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, short

Summary

CodeHunters presented their approach at SemEval-2026 Task 13, focusing on the detection of machine-generated code across various programming languages, code generators, and application scenarios. Their participation specifically involved Subtasks A and B of the competition. For these subtasks, the team fine-tuned three distinct pre-trained models: UniXCoder, CodeT5, and codeBERT. The paper, published in the Proceedings of the 20th International Workshop on Semantic Evaluation (2026) by the Association for Computational Linguistics, details the specific methodologies employed for both subtasks. This work, presented in San Diego, California, USA, on pages 270–276, contributes to the ongoing challenge of distinguishing human-written code from AI-generated code.

Key takeaway

For Machine Learning Engineers building systems to detect machine-generated code, you should consider fine-tuning pre-trained models such as UniXCoder, CodeT5, or codeBERT. This approach, demonstrated at SemEval-2026 Task 13, offers a robust method for distinguishing AI-written code across diverse programming languages and application scenarios. Your focus should be on adapting these models to specific subtasks to maximize detection accuracy and reliability.

Key insights

Fine-tuning pre-trained models like UniXCoder, CodeT5, and codeBERT effectively detects machine-generated code.

Principles

• Pre-trained models are adaptable for code detection.
• Fine-tuning enhances model performance on specific tasks.

Method

The method involved fine-tuning UniXCoder, CodeT5, and codeBERT, three pre-trained models, for detecting machine-generated code in various languages and scenarios.

In practice

• Evaluate UniXCoder, CodeT5, and codeBERT for code detection.
• Apply fine-tuning to pre-trained models for similar tasks.

Topics

• Machine-Generated Code Detection
• Code Language Models
• Fine-tuning
• UniXCoder
• CodeT5
• codeBERT
• SemEval-2026

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 10: Finetuning LLMs for Conspiracy Detection — Data augmentation and self-training can effectively finetune LLMs for specialized text classification tasks with limited data.
• 5 Free AI Tools to Understand Code and Generate Documentation — AI tools significantly streamline code comprehension and documentation generation for developers.
• Code Is More Than Text: Uncertainty Estimation for Code Generation — Code generation uncertainty estimation requires specialized methods accounting for code's unique properties, outperforming natural language-derived baselines.
• MOLOT System Card: Malicious Operational Logic Observation Transformer — Static behavior-sequence modeling offers accurate, explainable, and deployable malicious-code detection for modern DevSecOps.
• Detecting Functional Memorization in Code Language Models — Code LLMs can functionally memorize logic even when textually dissimilar, requiring advanced auditing.
• Training Agents: Live tutorial on how to fine-tune a coding agent for continual learning — The tutorial outlines a supervised fine-tuning (SFT) baseline for coding agents using public traces and TRL + LoRA.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.