# Team DACTYL at PAN 2026: Bayesian Data Mixing and Empirical X-risk Minimization for AI-text Detection

arXiv: https://arxiv.org/abs/2607.17382 · HTML: https://mlsanity.com/paper/2607.17382/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 27 of its 46 references that are in the map.
- First posted on arXiv 19 Jul 2026.

## Builds on (most cited first, 27 of 27)

- [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://mlsanity.com/paper/2006.03654/) (2020)
- [Weight Uncertainty in Neural Networks](https://mlsanity.com/paper/1505.05424/) (2015)
- [Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference](https://mlsanity.com/paper/2412.13663/) (2024)
- [Dataset Cartography: Mapping and Diagnosing Datasets with Training Dynamics](https://mlsanity.com/paper/2009.10795/) (2020)
- [Mercury: Ultra-Fast Language Models Based on Diffusion](https://mlsanity.com/paper/2506.17298/) (2025)
- [RAID: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors](https://mlsanity.com/paper/2405.07940/) (2024)
- [Well-Read Students Learn Better: On the Importance of Pre-training Compact Models](https://mlsanity.com/paper/1908.08962/) (2019)
- [Generalization in NLI: Ways (Not) To Go Beyond Simple Heuristics](https://mlsanity.com/paper/2110.01518/) (2021)
- [DetectRL: Benchmarking LLM-Generated Text Detection in Real-World Scenarios](https://mlsanity.com/paper/2410.23746/) (2024)
- [Technical Report on the Pangram AI-Generated Text Classifier](https://mlsanity.com/paper/2402.14873/) (2024)
- [When AUC meets DRO: Optimizing Partial AUC for Deep Learning with Non-Convex Convergence Guarantee](https://mlsanity.com/paper/2203.00176/) (2022)
- [Beemo: Benchmark of Expert-edited Machine-generated Outputs](https://mlsanity.com/paper/2411.04032/) (2024)
- [When More is Less: Incorporating Additional Datasets Can Hurt Performance By Introducing Spurious Correlations](https://mlsanity.com/paper/2308.04431/) (2023)
- [LibAUC: A Deep Learning Library for X-Risk Optimization](https://mlsanity.com/paper/2306.03065/) (2023)
- [Is Contrasting All You Need? Contrastive Learning for the Detection and Attribution of AI-generated Text](https://mlsanity.com/paper/2407.09364/) (2024)
- [Reliably Bounding False Positives: A Zero-Shot Machine-Generated Text Detection Framework via Multiscaled Conformal Prediction](https://mlsanity.com/paper/2505.05084/) (2025)
- [MovieSum: An Abstractive Summarization Dataset for Movie Screenplays](https://mlsanity.com/paper/2408.06281/) (2024)
- [MCGrad: Multicalibration at Web Scale](https://mlsanity.com/paper/2509.19884/) (2025)
- [Overview of PAN 2026: Voight-Kampff Generative AI Detection, Text Watermarking, Multi-Author Writing Style Analysis, Generative Plagiarism Detection, and Reasoning Trajectory Detection](https://mlsanity.com/paper/2602.09147/) (2026)
- [DACTYL: Diverse Adversarial Corpus of Texts Yielded from Large Language Models](https://mlsanity.com/paper/2508.00619/) (2025)
- [CoCoNUTS: Concentrating on Content while Neglecting Uninformative Textual Styles for AI-Generated Peer Review Detection](https://mlsanity.com/paper/2509.04460/) (2025)
- [A Comprehensive Dataset for Human vs. AI Generated Text Detection](https://mlsanity.com/paper/2510.22874/) (2025)
- [Mitigating Shortcut Learning with InterpoLated Learning](https://mlsanity.com/paper/2507.05527/) (2025)
- [LLMTrace: A Corpus for Classification and Fine-Grained Localization of AI-Written Text](https://mlsanity.com/paper/2509.21269/) (2025)
- [MAGA-Bench: Machine-Augment-Generated Text via Alignment Detection Benchmark](https://mlsanity.com/paper/2601.04633/) (2026)
- [On the Effectiveness of LLM-Specific Fine-Tuning for Detecting AI-Generated Text](https://mlsanity.com/paper/2601.20006/) (2026)
- [Reasoning-Aware AIGC Detection via Alignment and Reinforcement](https://mlsanity.com/paper/2604.19172/) (2026)

## Abstract (arXiv)

Existing research shows that AI-generated text detection classifiers achieve strong in-distribution (ID) performance but do not maintain the same performance on out-of-distribution (OOD) texts, suggesting overfitting to dataset-specific features. However, combining different training datasets doesn't always improve performance and, in some cases, can even encourage shortcut learning. To address this issue, we fine-tune BERT-tiny models with Bayesian classification heads to select texts across three different datasets to use as a consolidated training set. We trained three different classifiers: fine-tuned DeBERTa-V3-large and ModernBERT-large classifiers via empirical X-risk minimization, and an MCGrad model that calibrates the predictions from the ModernBERT-large classifier. The DeBERTa-V3-large-large classifier achieves a mean score of 0.882 on the PAN 2026 test set across five metrics: AUROC, F1, C@1, Brier score, and F0.5u. ModernBERT-large achieves a score of 0.96 while MCGrad achieves the best score of the three with a mean score of 0.974, ranking second on the leaderboard. Our results highlight that careful dataset curation can lead to strong OOD performance. We release our ModernBERT-large and DeBERTa-V3-large models at https://huggingface.co/collections/ShantanuT01/panclef-2026 .