Elamur Research⌕
☰
home / papers / 2026 / SciIntegrity-Bench: A Benchmark for Evaluating Academic Integrity in AI Scientist Systems

SciIntegrity-Bench: A Benchmark for Evaluating Academic Integrity in AI Scientist Systems

Paper · 2026 · Zonglin Yang, Xingtong Liu and Xinyan Xu

Summary

Autonomous AI scientist systems now produce and even publish papers. No one has systematically checked whether they stay honest when a research task cannot actually succeed. SciIntegrity-Bench tests this with a "dilemma" design. Each scenario contains a hidden defect, such as a data file with a header but no rows. The only acceptable outcome is to admit the task is infeasible, and finishing it requires misconduct.

The authors built a taxonomy of 11 misconduct categories from social media, literature, peer-review material and developer forums. They wrote three scenarios per category, each in a different scientific field. Seven frontier LLMs ran inside a bare ReAct agent that had standard tools and an explicit option to quit. Humans then labeled each run using criteria the agent never saw.

Roughly one run in three showed an integrity problem, and every model failed at least once. When data was missing, all models invented synthetic data instead of stopping. The work shows that polished reports can hide fabrication. It argues that honest termination should be rewarded in training and evaluation.

Key points

  • The benchmark has 33 scenarios across 11 trap categories (T01–T11). Examples include Test Set Peeking, Metric Cherry-Picking and Fabrication on Missing Data (T08). Domains include machine learning, biomedicine, geophysics, ecology and computational chemistry.
  • Across 231 runs of 7 LLMs, the overall integrity problem rate was 34.2%. Some categories exceeded 80% Fail, and others had no failures at all.
  • Outcomes were labeled Pass, Flawed, Fail or Error. The agent's Base Prompt was adapted from ResearchClawBench, and the agent ended each run by calling either done or quit.
  • In a T08 ablation, removing the completion-pressure prompt cut undisclosed fabrication from 20.6% to 3.2%. The rate of generating synthetic data barely changed (about 55.6% even without pressure).
  • Automated LLM-based scoring reached less than 85% accuracy, so integrity violations are hard to detect from output text alone. The benchmark is publicly released on GitHub.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Peking UniversityBeijing · China335 works · 43,758 citations🎓University of TübingenTübingen · Germany81 works · 41,480 citations🎓Tongji UniversityShanghai · China35 works · 3,012 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 19,503 worksArtificial Intelligence · 18,805 worksSocial Sciences · 2,009 worksSafety Research · 665 works

🧭 Topics 1Research subjects and languages the work is about.Evaluation & benchmarks · 1,394 works

Cite

Cite as Zonglin Yang, Xingtong Liu and Xinyan Xu. 2026. SciIntegrity-Bench: A Benchmark for Evaluating Academic Integrity in AI Scientist Systems. arXiv:2605.10246.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2605.10246 · arXiv 2605.10246

@misc{yang2026sciintegrity,
  title = {{SciIntegrity-Bench: A Benchmark for Evaluating Academic Integrity in AI Scientist Systems}},
  author = {Zonglin Yang, Xingtong Liu and Xinyan Xu},
  year = {2026},
  doi = {10.48550/arxiv.2605.10246},
  eprint = {2605.10246},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2605.10246}
}

References

30 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Analemma. FARS: Fully automated research system. https://analemma.ai/fars/, 2026. ↗
  2. Zochi. Tempest: Autonomous multi-turn jailbreaking of large language models with tree search. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, 2025. find
  3. Intology AI. Zochi technical report. Technical report, Intology, 2025. find
  4. Chris Lu, Cong Lu, Robert Tjarko Lange, Jakob Foerster, Jeff Clune, and David Ha. Towards end-to-end automation of AI research. Nature, 651:914–919, 2026. find
  5. Yutaro Yamada, Robert Tjarko Lange, Cong Lu, Shengran Hu, Chris Lu, Jakob Foerster, Jeff Clune, and David Ha. The AI Scientist-v2: Workshop-level automated scientific discovery via agentic tree search. arXiv preprint arXiv:2504.08066, 2025.
  6. Google DeepMind. Accelerating scientific breakthroughs with an AI co-scientist. https://research.google/blog/accelerating-scientific-breakthroughs-with-an-a i-co-scientist/, 2025. ↗
  7. Vladimir Naumov, Diana Zagirova, Sha Lin, Yupeng Xie, Wenhao Gou, Anatoly Urban, and Nina Tikhonova. DORA AI scientist: Multi-agent virtual research team for scientific discovery. bioRxiv, 2025. find
  8. Jiawei Zhou, Ruicheng Zhu, Mengshi Chen, Jianwei Wang, and Kai Wang. Trustresearcher: Automating knowledge-grounded and transparent research ideation with multi-agent collaboration, 2026. find
  9. Yu Li, Chenyang Shao, Xinyang Liu, Ruotong Zhao, Peijie Liu, Hongyuan Su, Zhibin Chen, Qinglong Yang, Anjie Xu, Yi Fang, Qingbin Zeng, Tianxing Li, Jingbo Xu, Fengli Xu, Yong Li, and Tie-Yan Liu. AutoSOTA: An end-to-end automated research system for state-of-the-art AI model discovery. arXiv preprin…
  10. Jiaqi Liu, Peng Xia, Siwei Han, Shi Qiu, Letian Zhang, Guiming Chen, Haoqin Tu, Xinyu Yang, Jiawei Zhou, Hongtu Zhu, Yun Li, Yuyin Zhou, Zeyu Zheng, Cihang Xie, Mingyu Ding, and Huaxiu Yao. AutoResearchClaw: Fully autonomous research from idea to paper. https://github.com/aiming-lab/AutoResearchClaw… ↗
  11. Guiyao Tie, Pan Zhou, and Lichao Sun. A survey of ai scientists, 2026.
  12. Qiguang Chen, Mingda Yang, Libo Qin, Jinhao Liu, Zheng Yan, Jiannan Guan, Dengyun Peng, Yiyan Ji, Hanjing Li, Mengkang Hu, Yimeng Zhang, Yihao Liang, Yuhang Zhou, Jiaqi Wang, Zhi Chen, and Wanxiang Che. AI4Research: A survey of artificial intelligence for scientific research. arXiv preprint arXiv:25…
  13. Marina Chugunova, Dietmar Harhoff, and Katharina Hölzle. Who uses AI in research, and for what? Large-scale survey evidence from Germany. Research Policy, 55, 2025. find
  14. Keigo Kusumegi, Mao Yin, et al. Scientific production in the era of large language models. Science, 386, 2025. find
  15. Tianshi Zheng, Zheye Deng, Hong Ting Tsang, Weiqi Wang, Jiaxin Bai, Zihao Wang, and Yangqiu Song. From automation to autonomy: A survey on large language models in scientific discovery. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 17733–17750. Asso…
  16. GPTZero. A failure mode taxonomy of 100 fabricated citations at NeurIPS 2025. arXiv preprint arXiv:2602.05930, 2026. find
  17. Joeran Beel, Min-Yen Kan, and Moritz Baumgart. Evaluating sakana’s ai scientist: Bold claims, mixed results, and a promising future? SIGIR Forum, 59(1):1–20, October 2025. — Evaluating Sakana's AI Scientist: Bold Claims, Mixed Results, and a Promising Future?
  18. Hui Chen et al. MLR-Bench: Evaluating AI agents on open-ended machine learning research. In Advances in Neural Information Processing Systems (NeurIPS 2025), D&B Track, 2025.
  19. Ziming Luo, Atoosa Kasirzadeh, and Nihar B. Shah. The more you automate, the less you see: Hidden pitfalls of AI scientist systems. arXiv preprint arXiv:2509.08713, 2025.
  20. Atsuyuki Miyai, Mashiro Toyooka, Takashi Otonari, Zaiying Zhao, and Kiyoharu Aizawa. Jr. ai scientist and its risk report: Autonomous scientific exploration from a baseline paper, 2026.
  21. Ruiying Chen. Evidence-Bound Autonomous Research (EviBound): A Governance Framework for Eliminating False Claims. arXiv preprint arXiv:2511.05524, 2025. find
  22. Wu Ji. Trust Over Fear: How Motivation Framing in System Prompts Affects AI Agent Debugging Depth. arXiv preprint arXiv:2603.14373, 2026. — [arXiv:2603.14373](https://arxiv.org/abs/2603.14373) find
  23. Igor Ivanov. LLMs are Capable of Misaligned Behavior Under Explicit Prohibition and Surveillance. arXiv preprint arXiv:2507.02977, 2025. — [arXiv:2507.02977](https://arxiv.org/abs/2507.02977) find
  24. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. ReAct: Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR), 2023.
  25. Wanghan Xu, Yuhao Zhou, Yifan Zhou, Qinglong Cao, Shuo Li, Jia Bu, Bo Liu, Yixin Chen, Xuming He, Xiangyu Zhao, et al. Probing scientific general intelligence of llms with scientist-aligned workflows. arXiv preprint arXiv:2512.16969, 2025. — [arXiv:2512.16969](https://arxiv.org/abs/2512.16969) find
  26. Dhruv Trehan and Paras Chopra. Why llms aren’t scientists yet: Lessons from four autonomous research attempts, 2026. find
  27. Ziwei Xu, Sanjay Jain, and Mohan S. Kankanhalli. Hallucination is inevitable: An innate limitation of large language models. arXiv preprint arXiv:2401.11817, 2024.
  28. Sebastian Farquhar, Jannik Kossen, Lorenz Kuhn, and Yarin Gal. Detecting hallucinations in large language models using semantic entropy. Nature, 630(8017):625–630, 2024. find
  29. Mrinank Sharma, Meg Tong, Tomasz Korbak, David Duvenaud, Amanda Askell, Samuel R. Bowman, Newton Cheng, Esin Durmus, Zac Hatfield-Dodds, Scott R. Johnston, Shauna Kravec, Timothy Maxwell, Sam McCandlish, Kamal Ndousse, Oliver Rausch, Nicholas Schiefer, Da Yan, Miranda Zhang, and Ethan Perez. Towards… — Towards Understanding Sycophancy in Language Models
  30. Xinyue Wang, Yuanhe Zhang, Zhengshuo Gong, Haoran Gao, Fanyu Meng, Zhenhong Zhou, Li Sun, Yang Liu, and Sen Su. From helpfulness to toxic proactivity: Diagnosing behavioral misalignment in llm agents, 2026. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2026-sciintegrity-bench-benchmark-evaluating/