https://gptzero.me/news/iclr-2026/ Toggle menu [solid-logo-2] * Search * Dashboard * All News * Education * Investigations * Technology * Pricing * About Us Search Search Dashboard Investigations Featured GPTZero finds over 50 new hallucinations in ICLR 2026 submissions GPTZero used our Citation Check tool to find 50+ hallucinations under review at ICLR, each of which were missed by 3-5 peer reviewers. [avat] Paul Esau [avat] [avat] Alex Cui Paul Esau, Nazar Shmatko, Alex Cui Dec 05, 2025 * 19 min read Share on X Share on Facebook Share on Linkedin Fact checked Copy citation to this article Copy link Send by email [ICLR_Logo] International Conference on Learning Representations is among the world's most prestigious machine learning research conferences. Table of contents Peer review is under siege. By speeding up the writing process, LLMs and other AI tools are overwhelming scholarly journals and conferences and the peer review pipeline with hallucinated papers ("AI slop"). These aren't just issues for low-ranking journals with high acceptance rates. The GPTZero team used our Citation Check tool to scan 300 papers under review by the prestigious International Conference on Learning Representations (ICLR). We discovered that 50 submissions included at least one obvious hallucitation, which were not previously reported. Worryingly, each of these submissions has already been reviewed by 3-5 peer experts, most of whom missed the fake citation(s). This failure suggests that some of these papers might have been accepted by ICLR without any intervention. Some had average ratings of 8/10, meaning they would almost certainly have been published. Is there a specific report or published article you think we should check? Submit Here Here's 50 confirmed hallucitations in ICLR 2026 submissions In the table below, we've included a specific human-verified hallucitation our tool flagged in each paper. According to the ICLR's editorial policy, even a single, clear hallucitation is an ethics violation that could lead to the paper's rejection. Given that we've only scanned 300 out of 20,000 submissions, we estimate that we will find 100s of hallucinated papers in the coming days. +--------------------------------------------------------------------------------------------------------------------------------------------------------------+ | |Average| | | | | |Title |Review |Paper Link |Citation Check Scan Link |Example of Verified Hallucination |Comment | | |Rating | | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |TamperTok: | |TamperTok: | | | | |Forensics-Driven | |Forensics-Driven | |Chong Zou, Zhipeng Wang, Ziyu Li, Nan Wu, Yuling Cai, | | |Tokenized | |Tokenized |https://app.gptzero.me/documents/ |Shan Shi, Jiawei Wei, Xia Sun, Jian Wang, and Yizhou |This paper | |Autoregressive |8.0 |Autoregressive |4645494f-70eb-40bb-aea7-0007e13f7179|Wang. Segment everything everywhere all at once. In |exists, but | |Framework for Image | |Framework for Image |/share |Advances in Neural Information Processing Systems |all authors | |Tampering | |Tampering | |(NeurIPS), volume 36, 2023. |are wrong. | |Localization | |Localization | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | | |The paper and| |MixtureVitae: Open | |MixtureVitae: Open | | |first 3 | |Web-Scale Pretraining| |Web-Scale Pretraining| |Dan Hendrycks, Collin Burns, Steven Basart, Andy |authors | |Dataset With High | |Dataset With High |https://app.gptzero.me/documents/ |Critch, Jerry Li, Dawn Ippolito, Aina Lapedriza, |match. The | |Quality Instruction |8.0 |Quality Instruction |bfd10666-ea2d-454c-9ab2-75faa8b84281|Florian Tramer, Rylan Macfarlane, Eric Jiang, et al. |last 7 | |and Reasoning Data | |and Reasoning Data |/share |Measuring massive multitask language understanding. In |authors are | |Built from Permissive| |Built from Permissive| |Proceedings of the International Conference on Learning|not on the | |Text Sources | |Text Sources | | |Representations (ICLR), 2021. |paper, and | | | |OpenReview | | |some of them | | | | | | |do not exist | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Catch-Only-One: | |Catch-Only-One: | | | | |Non-Transferable | |Non-Transferable | https://app.gptzero.me/documents/ |Dinghuai Zhang, Yang Song, Inderjit Dhillon, and Eric | | |Examples for |6.0 |Examples for |9afb1d51-c5c8-48f2-9b75-250d95062521|Xing. Defense against adversarial attacks using |No Match | |Model-Specific | |Model-Specific |/share |spectral regularization. In International Conference on| | |Authorization | |Authorization | | |Learning Representations (ICLR), 2020. | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |OrtSAE: Orthogonal | |OrtSAE: Orthogonal | https://app.gptzero.me/documents/ |Robert Huben, Logan Riggs, Aidan Ewart, Hoagy |This paper | |Sparse Autoencoders |6.0 |Sparse Autoencoders |e3f155d7-067a-4720-adf8-65dc9dc714b9|Cunningham, and Lee Sharkey. Sparse autoencoders can |exists, but | |Uncover Atomic | |Uncover Atomic |/share |interpret randomly initialized transformers, 2025. URL |all authors | |Features | |Features | OpenReview| |https://arxiv.org/ abs/2501.17727. |are wrong. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | |Principled Policy | |David Rein, Stas Gaskin, Lajanugen Logeswaran, Adva | | |Principled Policy | |Optimization for LLMs| https://app.gptzero.me/documents/ |Wolf, Oded teht sun, Jackson H. He, Divyansh Kaushik, |All authors | |Optimization for LLMs|5.0 |via Self-Normalized |54c8aa45-c97d-48fc-b9d0-d491d54df8d3|Chitta Baral, Yair Carmon, Vered Shwartz, Sang-Woo Lee,|except the | |via Self-Normalized | |Importance Sampling ||/share |Yoav Goldberg, C. J. H. un, Swaroop Mishra, and Daniel |first are | |Importance Sampling | |OpenReview | |Khashabi. Gpqa: A graduate-level google-proof q\&a |fabricated. | | | | | |benchmark, 2023 | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | | |Authors and | | | | | |Andrew Chen, Andy Chow, Aaron Davidson, Arjun DCunha, |conference | |PDMBench: A | |PDMBench: A | |Ali Ghodsi, Sue Ann Hong, Andy Konwinski, Clemens |match this | |Standardized Platform| |Standardized Platform| https://app.gptzero.me/documents/ |Mewald, Siddharth Murching, Tomas Nykodym, et al. |paper, but | |for Predictive |4.5 |for Predictive |5c55afe7-1689-480d-ac44-9502dc0f9229|Mlflow: A platform for managing the machine learning |title is | |Maintenance Research | |Maintenance Research |/share |lifecycle. In Proceedings of the Fourth International |somewhat | | | || OpenReview | |Workshop on Data Management for End-to-End Machine |different and| | | | | |Learning, pp. 1-4. ACM, 2018. |the year is | | | | | | |wrong. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | |IMPQ: | | |The arXiv ID | |IMPQ: | |Interaction-Aware | | |is real, but | |Interaction-Aware | |Layerwise Mixed | https://app.gptzero.me/documents/ |Chen Zhu et al. A survey on efficient deployment of |the paper has| |Layerwise Mixed |4.5 |Precision |5461eefd-891e-4100-ba1c-e5419af520c0|large language models. arXiv preprint arXiv:2307.03744,|different | |Precision | |Quantization for LLMs|/share |2023. |authors and a| |Quantization for LLMs| || OpenReview | | |different | | | | | | |title. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |C3-OWD: A Curriculum | |C3-OWD: A Curriculum | | | | |Cross-modal | |Cross-modal | https://app.gptzero.me/documents/ |K. Marino, R. Salakhutdinov, and A. Gupta. Fine-grained|Authors and | |Contrastive Learning |4.5 |Contrastive Learning |c07521cd-2757-40a2-8dc1-41382d7eb11b|image classification with learnable semantic parts. In |subject match| |Framework for | |Framework for |/share |Proceedings of the IEEE/CVF Conference on Computer |this paper | |Open-World Detection | |Open-World Detection | |Vision and Pattern Recognition, pp. 4500-4509, 2019. | | | | || OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |TopoMHC: | |TopoMHC: | |Yuchen Han, Yohan Kim, Dalibor Petrovic, Alessandro | | |Sequence-Topology | |Sequence-Topology | https://app.gptzero.me/documents/ |Sette, Morten Nielsen, and Bjoern Peters. Deepligand: a| | |Fusion for MHC |4.5 |Fusion for MHC |8da4f86c-00d8-4d73-81dd-c168c0bfdf4e|deep learning framework for peptide-mhc binding |No Match | |Binding | |Binding | OpenReview |/share |prediction. Bioinformatics, 39 (1):btac834, 2023. doi: | | | | | | |10.1093/bioinformatics/btac834. | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | |Yugandhar Balaji, Jianwei Yang, Zhen Xu, Menglei Chai, |This paper | |Can Text-to-Video | |Can Text-to-Video | |Zhoutong Xu, Ersin Yumer, Greg Shakhnarovich, and Deva |exists, but | |Models Generate | |Models Generate | https://app.gptzero.me/documents/ |Ramanan. Conditional gan with discriminative filter |the authors | |Realistic Human |4.5 |Realistic Human |f52aad2d-2253-44bf-80ba-8e8668df650f|generation for text-to-video synthesis. In Proceedings |and page | |Motion? | |Motion? | OpenReview |/share |of the 28th International Joint Conference on |numbers are | | | | | |Artificial Intelligence (IJCAI), pp. 2155-2161, July |wrong. | | | | | |2019. doi: 10.24963/ijcai.2019/276. | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |GRF-LLM: | |GRF-LLM: | | | | |Environment-Aware | |Environment-Aware | |Junting Chen, Yong Zeng, and Rui Zhang. Rfcanvas: A |Title | |Wireless Channel | |Wireless Channel | https://app.gptzero.me/documents/ |radio frequency canvas for wireless network design. In |partially | |Modeling via |4.0 |Modeling via |c3e66b9c-20b4-4c50-b881-e40aba2a514f|IEEE International Conference on Communications, pp. |matches this | |LLM-Guided 3D | |LLM-Guided 3D |/share |1-6, 2024. |article. | |Gaussians | |Gaussians | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Listwise Generalized | |Listwise Generalized | | | | |Preference | |Preference | https://app.gptzero.me/documents/ |Kaixuan Zhou, Jiaqi Liu, Yiding Wang, and James Zou. | | |Optimization with |4.0 |Optimization with |bbeecf1c-189a-4311-999b-617aab686ea9|Generalized direct preference optimization. arXiv |No Match | |Process-aware Signals| |Process-aware Signals|/share |preprint arXiv:2402.05015, 2024. | | |for LLM Reasoning | |for LLM Reasoning | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | |IUT-Plug: A Plug-in | |Yash Goyal, Anamay Mohapatra, Nihar Kwatra, and Pawan |This paper | |IUT-Plug: A Plug-in | |tool for Interleaved |https://app.gptzero.me/documents/ |Goyal. A benchmark for compositional text-to-image |exists, but | |tool for Interleaved |4.0 |Image-Text Generation|0f12d2fc-403b-4859-8d00-f75fd9f56e39|synthesis. In Thirty-fifth Conference on Neural |the authors | |Image-Text Generation| || OpenReview |/share |Information Processing Systems Datasets and Benchmarks |are all | | | | | |Track (Round 1), 2021. |wrong. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Resolving the | |Resolving the | | |No match; | |Security-Auditability| |Security-Auditability| https://app.gptzero.me/documents/ |Yixiang Ma, Ziyi Liu, Zhaoyu Wang, Zhaofeng Xu, Yitao |although this| |Dilemma with |4.0 |Dilemma with |5cee5c3a-5e75-4063-a054-1e934a071705|Wang, and Yang Liu. Safechain: A framework for securely|paper is | |Auditable Latent | |Auditable Latent |/share |executing complex commands using large language models.|closely | |Chain-of-Thought | |Chain-of-Thought | | |arXiv preprint arXiv:2402.16521, 2024a. |related. | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |ThinkGeo: Evaluating | |ThinkGeo: Evaluating | https://app.gptzero.me/documents/ |Yunzhu Yang, Shuang Li, and Jiajun Wu. MM-ReAct: | | |Tool-Augmented Agents|4.0 |Tool-Augmented Agents|f3441445-5401-48e9-9617-09a635992ff9|Prompting chatgpt to multi-modal chain-ofthought |No Match | |for Remote Sensing | |for Remote Sensing |/share |reasoning. arXiv preprint arXiv:2401.04740, 2024. | | |Tasks | |Tasks | OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Taming the Judge: | |Taming the Judge: | |Chenglong Wang, Yang Liu, Zhihong Xu, Ruochen Zhang, |All authors | |Deconflicting AI | |Deconflicting AI | https://app.gptzero.me/documents/ |Jiahao Wu, Tao Luo, Jingang Li, Xunliang Liu, Weiran |except the | |Feedback for Stable |3.5 |Feedback for Stable |80c64df2-eee6-41aa-90cc-3f835b128747|Qi, Yujiu Yang, et al. Gram-r ${ }^{8}$ : Self-training|first are | |Reinforcement | |Reinforcement |/share |generative foundation reward models for reward |fabricated | |Learning | |Learning | OpenReview| |reasoning. arXiv preprint arXiv:2509.02492, 2025b. |and the title| | | | | | |is altered. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | | |Two papers | | | |DANCE-ST: Why | | |with similar | |DANCE-ST: Why | |Trustworthy AI Needs | |Sardar Asif, Saad Ghayas, Waqar Ahmad, and Faisal |titles exist | |Trustworthy AI Needs | |Constraint Guidance, | https://app.gptzero.me/documents/ |Aadil. Atcn: an attention-based temporal convolutional |here and here| |Constraint Guidance, |3.5 |Not Constraint |3ebd71b4-560d-4fa3-a0d3-ed2fa13c519f|network for remaining useful life prediction. The |, but the | |Not Constraint | |Penalties | |/share |Journal of Supercomputing, 78(1): $1-19,2022$. |authors, | |Penalties | |OpenReview | | |journal, and | | | | | | |date do not | | | | | | |match. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Federated | |Federated | | | | |Hierarchical | |Hierarchical | | | | |Anti-Forgetting | |Anti-Forgetting | https://app.gptzero.me/documents/ |Arslan Chaudhry, Arun Mallya, and Abhinav Srivastava. | | |Framework for |3.33 |Framework for |ae10437b-c65b-455b-ad22-918742a5ed82|Fedclassil: A benchmark for classincremental federated |No Match | |Class-Incremental | |Class-Incremental |/share |learning. In NeurIPS, 2023. | | |Learning with Large | |Learning with Large | | | | |Pre-Trained Models | |Pre-Trained Models | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Chain-of-Influence: | |Chain-of-Influence: | | | | |Tracing | |Tracing | | | | |Interdependencies | |Interdependencies | https://app.gptzero.me/documents/ |Ishita et al. Bardhan. Icu length-of-stay prediction | | |Across Time and |3.33 |Across Time and |dff2c063-6986-4241-8c20-4327a39d4d4b|with interaction-based explanations. Journal of |No Match | |Features in Clinical | |Features in Clinical |/share |Biomedical Informatics, 144:104490, 2024. | | |Predictive Modeling | |Predictive Modeling || | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |TRACEALIGN - Tracing | |TRACEALIGN - Tracing | | |This article | |the Drift: | |the Drift: | | |is similar, | |Attributing Alignment| |Attributing Alignment| https://app.gptzero.me/documents/ |Lisa Feldman Barrett. Emotions are constructed: How |but the | |Failures to |3.33 |Failures to |4b379aba-8d8a-427b-ac67-d13af5eda8c9|brains make meaning. Current Directions in |title, and | |Training-Time Belief | |Training-Time Belief |/share |Psychological Science, 25(6):403-408, 2016. |metadata are | |Sources in LLMs | |Sources in LLMs | | | |different. | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |MEMORIA: A Large | |MEMORIA: A Large | | | | |Language Model, | |Language Model, | |Yang Cao, Rosa Martinez, and Sarah Thompson. Preserving| | |Instruction Data and | |Instruction Data and | https://app.gptzero.me/documents/ |indigenous languages through neural language models: | | |Evaluation Benchmark |3.33 |Evaluation Benchmark |956129a3-11ee-4503-92e3-3ed5db12d2d6|Challenges and opportunities. Computational |No Match | |for Intangible | |for Intangible |/share |Linguistics, 49(3):567-592, 2023. | | |Cultural Heritage | |Cultural Heritage | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Reflexion: Language | |Reflexion: Language | | | | |Models that Think | |Models that Think |https://app.gptzero.me/documents/ |Guang-He Xiao, Haolin Wang, and Yong-Feng Zhang. | | |Twice for |3.2 |Twice for |45f2f68d-df09-4bbf-8513-588fe24f26fa|Rethinking uncertainty in llms: A case study on a |No Match | |Internalized | |Internalized |/share |fact-checking benchmark. arXiv preprint | | |Self-Correction | |Self-Correction | | |arXiv:2305.11382, 2023. | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | |ECAM: Enhancing | |Atticus Geiger, Zhengxuan Wu, Yonatan Rozner, Mirac |A paper with | |ECAM: Enhancing | |Causal Reasoning in | |Suzgun Naveh, Anna Nagarajan, Jure Leskovec, |this title | |Causal Reasoning in | |Foundation Models | https://app.gptzero.me/documents/ |Christopher Potts, and Noah D Goodman. Causal |exists at the| |Foundation Models |3.0 |with Endogenous |d99a5552-38e0-459b-8746-4e64069b0640|interpretation of self-attention in pre-trained |given URL, | |with Endogenous | |Causal Attention |/share |transformers. In Advances in Neural Information |but the | |Causal Attention | |Mechanism | | |Processing Systems 36 (NeurIPS 2023), 2023. URL https:/|authors don't| |Mechanism | |OpenReview | |/proceedings.neurips.cc/paper_files/paper/ 2023/file/ |match. | | | | | |642a321fba8a0f03765318e629cb93ea-Paper-Conference.pdf. | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |MANTA: Cross-Modal | |MANTA: Cross-Modal | | |An article | |Semantic Alignment | |Semantic Alignment | | |with this | |and | |and | https://app.gptzero.me/documents/ |Guy Dove. Language as a cognitive tool to imagine goals|title exists,| |Information-Theoretic|3.0 |Information-Theoretic|381ed9a6-b168-4cd0-81ad-1f50139c0737|in curiosity-driven exploration. Nature Communications,|but author | |Optimization for | |Optimization for |/share |13(1):1-14, 2022. |and | |Long-form Multimodal | |Long-form Multimodal | | |publication | |Understanding | |Understanding | | | |don't match. | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |LOSI: Improving | |LOSI: Improving | | | | |Multi-agent | |Multi-agent | |Jing Liang, Fan Zhou, Shuying Li, Jun Chen, Guandong | | |Reinforcement | |Reinforcement | https://app.gptzero.me/documents/ |Zhou, Huaiming Xu, and Xin Li. Learning opponent | | |Learning via Latent |3.0 |Learning via Latent |53e86e4b-a7e2-48d0-976b-240bfc412836|behavior for robust cooperation in multi-agent |No Match | |Opponent Strategy | |Opponent Strategy |/share |reinforcement learning. IEEE Transactions on | | |Identification | |Identification | | |Cybernetics, 53(12):7527-7540, 2023. | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |The Dynamic | |The Dynamic | | | | |Interaction Field | |Interaction Field | |Kaj Bostrom and Greg Durrett. Byte-level representation| | |Transformer: A | |Transformer: A | https://app.gptzero.me/documents/ |learning for multi-lingual named entity recognition. | | |Universal, |3.0 |Universal, |80fd90a6-c99e-4c31-af72-0da9e90949f6|Proceedings of the 2020 Conference on Empirical Methods|No Match | |Tokenizer-Free | |Tokenizer-Free |/share |in Natural Language Processing (EMNLP), pp. 4617-4627, | | |Language Architecture| |Language Architecture| |2020. | | | | || OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Strategema: | |Strategema: | | | | |Probabilistic | |Probabilistic | |Tom Eccles, Jeffrey Tweedale, and Yvette Izza. Let's | | |Analysis of | |Analysis of | https://app.gptzero.me/documents/ |pretend: A study of negotiation with autonomous agents.| | |Adversarial |3.0 |Adversarial |1155e8a8-f679-4942-8fd9-c47fb64ad967|In 2009 IEEE/WIC/ACM International Joint Conference on |No Match | |Multi-Agent Behavior | |Multi-Agent Behavior |/share |Web Intelligence and Intelligent Agent Technology | | |with LLMs in Social | |with LLMs in Social | |(WI-IAT), volume 3, pp. 449-452. IEEE, 2009. | | |Deduction Games | |Deduction Games | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Understanding | |Understanding | | | | |Transformer | |Transformer | | | | |Architecture through | |Architecture through | https://app.gptzero.me/documents/ |Zijie J Wang, Yuhao Choi, and Dongyeop Wei. On the | | |Continuous Dynamics: |3.0 |Continuous Dynamics: |460a1a23-1a97-482a-9759-ade855a4a0b4|identity of the representation learned by pre-trained |No Match | |A Partial | |A Partial |/share |language models. arXiv preprint arXiv:2109.01819, 2021.| | |Differential Equation| |Differential Equation| | | | |Perspective | |Perspective | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | | |A similar | | | | | | |paper with | | | | | | |two matching | | | |Diffusion Aligned | https://app.gptzero.me/documents/ |Yujia Wang, Hu Huang, Cynthia Rudin, and Yaron |authors | |Diffusion Aligned |2.8 |Embeddings | |3d95a003-06c6-4233-881b-03b1e29b4ba2|Shaposhnik. Pacmap: Dimension reduction using pairwise |exists, but | |Embeddings | |OpenReview |/share |controlled manifold approximation projection. Machine |the other | | | | | |Learning, 110:559-590, 2021. |authors, | | | | | | |title, and | | | | | | |journal are | | | | | | |wrong. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Leveraging NLLB for | |Leveraging NLLB for | |Atnafa L. Tonja, Gebremedhin Gebremeskel, and Seid M. | | |Low-Resource | |Low-Resource | https://app.gptzero.me/documents/ |Yimam. Evaluating machine translation systems for | | |Bidirectional Amharic|2.5 |Bidirectional Amharic|813da6e2-f7e8-4c95-bdd8-7d29b8e4b641|ethiopian languages: A case study of amharic and afan |No Match | |- Afan Oromo Machine | |- Afan Oromo Machine |/share |oromo. Journal of Natural Language Engineering, 29 | | |Translation | |Translation | Open | |(3):456-478, 2023. | | | | |Review | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Certified Robustness | |Certified Robustness | |Huan Zhang, Hongge Chen, Chaowei Xiao, and Bo Zhang. | | |Training: Closed-Form| |Training: Closed-Form| https://app.gptzero.me/documents/ |Towards deeper and better certified defenses against | | |Certificates via |2.5 |Certificates via |53b60ef5-2ebf-403e-8123-3a9bb2da0f33|adversarial attacks. In International Conference on |No Match | |CROWN | |CROWN | OpenReview |/share |Learning Representations, 2019. URL https:// | | | | | | |openreview.net/forum?id=rJgG92A2m | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | |Context-Aware Input | | |Partial match| |Context-Aware Input | |Switching in Mobile | |Ishan Tarunesh, Syama Sundar Picked, Sai Krishna Bhat, |to this | |Switching in Mobile | |Devices: A | https://app.gptzero.me/documents/ |and Monojit Choudhury. Machine translation for |article, but | |Devices: A |2.5 |Multi-Language, |68998766-49c3-4269-9eca-3b6a76ed68b4|code-switching: A systematic literature review. In |authors, | |Multi-Language, | |Emoji-Integrated |/share |Proceedings of the 59th Annual Meeting of the |title, and | |Emoji-Integrated | |Typing System | | |Association for Computational Linguistics, pp. |metadata is | |Typing System | |OpenReview | |3654-3670, 2021. |largely | | | | | | |wrong. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | | |A paper with | | | |Five-Mode Tucker-LoRA| | |the same | |Five-Mode Tucker-LoRA| |for Video Diffusion | https://app.gptzero.me/documents/ |Shengming Chen, Yuxin Wang, et al. Videocrafter: Open |title exists,| |for Video Diffusion |2.5 |on Conv3D Backbones ||eb0fd660-ed00-4769-a940-3d093d4f1ec1|diffusion models for high-quality video generation. |but the | |on Conv3D Backbones | |OpenReview |/share |arXiv preprint arXiv:2305.07932, 2023b. |authors and | | | | | | |arXiv ID are | | | | | | |wrong. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Activation-Guided | |Activation-Guided | | | | |Regularization: | |Regularization: | | | | |Improving Deep | |Improving Deep | https://app.gptzero.me/documents/ |Wentao Cheng and Tong Zhang. Improving deep learning | | |Classifiers using |2.5 |Classifiers using |4031111e-24ef-4e06-908e-18ab99b08932|for classification with unknown label noise. In |A similar | |Feature-Space | |Feature-Space |/share |International Conference on Machine Learning, pp. |paper exists.| |Regularization with | |Regularization with | |6059-6081. PMLR, 2023. | | |Dynamic Prototypes | |Dynamic Prototypes | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Sparse-Smooth | |Sparse-Smooth | | | | |Decomposition for | |Decomposition for | https://app.gptzero.me/documents/ |Yutian Chen, Kun Zhang, Jonas Peters, and Bernhard | | |Nonlinear Industrial |2.5 |Nonlinear Industrial |c01ad49e-a788-4916-a6ee-f43314d14676|Scholkopf. Causal discovery and inference for |No Match | |Time Series | |Time Series |/share |nonstationary systems. Journal of Machine Learning | | |Forecasting | |Forecasting | | |Research, 22(103):1-72, 2021. | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | | |The paper | | | | | | |exists and | | | |PDE-Transformer: A | |Xuechen Li, Juntang Zhuang, Yifan Ding, Zhaozong Jin, |the first | |PDE-Transformer: A | |Continuous Dynamical | https://app.gptzero.me/documents/ |Yun chen Chen, and Stefanie Jegelka. Scalable gradients|author is | |Continuous Dynamical |2.0 |Systems Approach to |ba257eea-e86c-4276-84c0-08b7465e1e3e|for stochastic differential equations. In Proceedings |correct but | |Systems Approach to | |Sequence Modeling | |/share |of the Twenty Third International Conference on |all other | |Sequence Modeling | |OpenReview | |Artificial Intelligence and Statistics (AISTATS 2020), |authors and | | | | | |volume 108 of Proceedings of Machine Learning Research,|the page | | | | | |pp. 3898-3908, 2020. |range are | | | | | | |wrong | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |SAFE-LLM: A Unified | |SAFE-LLM: A Unified | | |A similar | |Framework for | |Framework for | https://app.gptzero.me/documents/ | |paper with | |Reliable, Safe, And |2.0 |Reliable, Safe, And |05ee7ff4-40e2-48b7-b5bd-8c307d7db669|Kuhn, J., et al. Semantic Entropy for Hallucination |different | |Secure Evaluation of | |Secure Evaluation of |/share |Detection. ACL 2023. |authors can | |Large Language Models| |Large Language Models| | |be found here| | | || OpenReview | | |. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |PIPA: An Agent for | |PIPA: An Agent for | |Alex Brown et al. Autonomous scientific experimentation| | |Protein Interaction | |Protein Interaction | https://app.gptzero.me/documents/ |at the advanced light source using | | |Identification and |2.0 |Identification and |5031a806-1271-4fd3-b333-2554f47cb9fa|language-model-driven agents. Nature Communications, |No Match | |Perturbation Analysis| |Perturbation Analysis|/share |16:7001, 2025. | | | | || OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | |Typed | |DeepMind. Gemma scope: Scaling mechanistic |ThA similar | |Typed | |Chain-of-Thought: A | |interpretability to chain of thought. DeepMind Safety |URL exists, | |Chain-of-Thought: A | |Curry-Howard | https://app.gptzero.me/documents/ |Blog, 2025. URL https:// |and the title| |Curry-Howard |2.0 |Framework for |9d2e3239-99db-4712-be7f-e032156d92a5|deepmindsafetyresearch.medium.com/ |is similar to| |Framework for | |Verifying LLM |/share |evaluating-and-monitoring-for-ai-scheming-8a7f2ce087f9.|this blog. | |Verifying LLM | |Reasoning | | |Discusses scaling mechanistic interpretability |However, no | |Reasoning | |OpenReview | |techniques to chain-of-thought and applications such as|exact match | | | | | |hallucination detection. |exists. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Graph-Based Operator | |Graph-Based Operator | https://app.gptzero.me/documents/ |Liu, Y., Lutjens, B., Azizzadenesheli, K., and | | |Learning from Limited|2.0 |Learning from Limited|6c52217f-fb88-4bd8-85aa-bd546e1fa88c|Anandkumar, A. (2022). U-netformer: A u-net style |No Match | |Data on Irregular | |Data on Irregular |/share |transformer for solving pdes. arXiv preprint | | |Domains | |Domains | OpenReview | |arXiv:2206.11832. | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |KARMA: | |KARMA: | |Reinaldo A. C. Bianchi, Luis A. Celiberto Jr, and Ramon| | |Knowledge-Aware | |Knowledge-Aware | https://app.gptzero.me/documents/ |Lopez de Mantaras. Knowledge-based reinforcement | | |Reward Mechanism |2.0 |Reward Mechanism |92b6492c-68ad-41a3-ae35-628d67f053e0|learning: A survey. Journal of Artificial Intelligence |No Match | |Adjustment via Causal| |Adjustment via Causal|/share |Research, 62:215-261, 2018. | | |AI | |AI | OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | | |the arXiv ID | | | |Microarchitecture Is | | |corresponds | |Microarchitecture Is | |Destiny: Performance | | |with a very | |Destiny: Performance | |and Accuracy of | https://app.gptzero.me/documents/ |Zhihang Jiang, Dingkang Wang, Yao Li, et al. Fp6-llm: |similar paper| |and Accuracy of |2.0 |Quantized LLMs on |4504a39a-af72-41ab-9679-6f6a017a3275|Efficient llm serving through fp6-centric co-design. |, but the | |Quantized LLMs on | |Consumer Hardware | |/share |arXiv preprint arXiv:2401.14112, 2024. |authors are | |Consumer Hardware | |OpenReview | | |wrong and the| | | | | | |title is | | | | | | |altered. | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Decoupling of | |Decoupling of | | | | |Experts: A | |Experts: A | https://app.gptzero.me/documents/ |H Zhang, Y L, X W, Y Z, X Z, H W, X H, K G, Z W, H W, H|No Match; | |Knowledge-Driven |1.6 |Knowledge-Driven |74eade70-da36-4635-8749-5e1d04748b6d|C, H L, and J W. Matrix data pile: A |arxiv is is | |Architecture for | |Architecture for |/share |trillion-tokenscale datasets for llm pre-training. |unrelated | |Efficient LLMs | |Efficient LLMs | | |arXiv preprint arXiv:2408.12151, 2024. | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |QUART: Agentic | |QUART: Agentic | | | | |Reasoning To Discover| |Reasoning To Discover| https://app.gptzero.me/documents/ |Meera Jain and Albert Chen. Explainable ai techniques | | |Missing Knowledge in |1.5 |Missing Knowledge in |c6f30343-3948-4c07-b7de-6b1407d5daa6|for medical applications: A comprehensive review. AI in|No Match | |Multi-Domain Temporal| |Multi-Domain Temporal|/share |Healthcare, 5:22-37, 2024. | | |Data. | |Data. | OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |From Physics-Informed| |From Physics-Informed| | | | |Models to Deep | |Models to Deep | | |The title | |Learning: | |Learning: | https://app.gptzero.me/documents/ | |matches this | |Reproducible AI |1.5 |Reproducible AI |a7ed6c42-4349-4b45-a356-0e325090e5af|MIT Climate Group. A cautionary tale for deep learning |paper, but | |Frameworks for | |Frameworks for |/share |in climate science. https://example. com, 2019. |the citation | |Climate Resilience | |Climate Resilience | | |is obviously | |and Policy Alignment | |and Policy Alignment | | |hallucinated.| | | || OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | |Roy Bar-Haim, Shachar Bhattacharya, Michal Jacovi, Yosi| | | | | | |Mass, Matan Orbach, Eyal Sliwowicz, and Noam Slonim. |A paper with | | | | | |Key point analysis via contrastive learning and |the same | |A superpersuasive | |A superpersuasive | https://app.gptzero.me/documents/ |extractive argument summarization. In Proceedings of |title exists,| |autonomous policy |1.5 |autonomous policy |b792a4de-baa8-47d4-b880-87b330a482ce|the 2021 Conference on Empirical Methods in Natural |but the | |debating system | |debating system | |/share |Language Processing, pages 7953-7962, Online and Punta |authors and | | | |OpenReview | |Cana, Dominican Republic, November 2021a. Association |URL are | | | | | |for Computational Linguistics. doi: 10.18653/v1/ |wrong. | | | | | |2021.emnlp-main.629. URL https://aclanthology.org/ | | | | | | |2021.emnlp-main. 629. | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |AnveshanaAI: A | |AnveshanaAI: A | | | | |Multimodal Platform | |Multimodal Platform | | | | |for Adaptive AI/ML | |for Adaptive AI/ML | | | | |Education Through | |Education Through | https://app.gptzero.me/documents/ |Shiyang Liu, Hongyi Xu, and Min Chen. Measuring and | | |Automated Question |1.5 |Automated Question |720d6d24-2223-4e0e-95b9-6dfce674f8c7|reducing perplexity in large-scale llms. arXiv preprint|No Match | |Generation and | |Generation and |/share |arXiv:2309.12345, 2023. | | |Interactive | |Interactive | | | | |Assessment | |Assessment | | | | | | | |OpenReview | | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |AI-Assisted Medical | |AI-Assisted Medical | https://app.gptzero.me/documents/ |[3] K. Arnold, J. Smith, and A. Doe. Variability in | | |Triage Assistant |1.0 |Triage Assistant | |391b5d76-929a-4f3f-addf-31f6993726f2|triage decision making. Resuscitation, 85:12341239, |No Match | | | |OpenReview |/share |2014. | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| |Deciphering | |Deciphering | | |A paper with | |Cross-Modal Feature | |Cross-Modal Feature | |Shuyang Basu, Sachin Y Gadre, Ameet Talwalkar, and Zico|this title | |Interactions in | |Interactions in | https://app.gptzero.me/documents/ |Kolter. Understanding multimodal llms: the mechanistic |exists, but | |Multimodal AIGC |0.67 |Multimodal AIGC |d4102812-01c4-45b2-aea8-59e467d31fd4|interpretability of llava in visual question answering.|the authors | |Models: A Mechanistic| |Models: A Mechanistic|/share |arXiv preprint arXiv:2411.17346, 2024. |and arXiv ID | |Interpretability | |Interpretability | | |are wrong. | |Approach | |Approach | OpenReview| | | | |---------------------+-------+---------------------+------------------------------------+-------------------------------------------------------+-------------| | | | | | |There is no | |Scalable Generative | |Scalable Generative | | |match for the| |Modeling of Protein | |Modeling of Protein | https://app.gptzero.me/documents/ |E. Brini, G. Jayachandran, and M. Karplus. |title and | |Ligand Trajectories |0.5 |Ligand Trajectories |32d43311-6e69-4b88-be99-682e4eb0c2cc|Coarse-graining biomolecular simulations via |authors, but | |via Graph Neural | |via Graph Neural |/share |statistical learning. J. Chem. Phys., 154:040901, 2021.|the journal, | |Diffusion Networks | |Diffusion Networks | | | |volume, and | | | |OpenReview | | |year match | | | | | | |this article | +--------------------------------------------------------------------------------------------------------------------------------------------------------------+ We'd like to acknowledge Siqiao Mu (Northwestern University), Alex Meiberg (University of Waterloo and the Perimeter Institute), and Winter Pearson (Caltech) for their contributions to this analysis. Under Siege Scientific journals and academic conferences are being crushed by an avalanche of submissions fueled by generative AI, paper mills, and publication pressure. Between 2016 and 2024, the number of scientific articles published annually jumped 48%, while retractions and other scandals proliferated. Many scientific conferences and journals are struggling to find qualified peer reviewers, while reviewers are " overwhelmed" by the increasing demand placed on their time. Academic conferences like ICLR are also under pressure. ICLR is one of the most important annual gatherings of artificial intelligence researchers on the planet, yet many recent conference submissions and peer reviews show signs of AI authorship. These signs run the spectrum from verbosity and excessive bullet points to fake data and "hallucitations." What Citation Check Surfaced Since GPTZero launched our Citation Check tool to catch hallucitations in January, we've tested it on RFK Jr.'s "MAHA" report , a scandal-ridden Deloitte Australia report, and hundreds of other documents. This week we used Citation Check to scan a sample set of 300 ICLR papers submitted to OpenReview. Our tool flagged 90 papers as containing at least one citation that appeared to not exist online. Following human verification, we determined that 50 papers included at least one actual hallucitation. Defining Hallucitations Given the high stakes for both researchers and editors, Citation Check is engineered to prioritize accuracy, provide transparency into the evaluation of each source, and err on the side of caution. It uses our AI agent, trained in-house, to flag any citations in a document that can't be found online. These flagged citations are not automatically hallucinations -- many archival documents or unpublished works can't be matched to an online source -- but they indicate which sources require further human scrutiny. Like ICLR, GPTZero recommends that a human determine if a flawed citation is an AI-generated fake or the result of a more conventional error. Although the line can be blurry, we define hallucitations as citations resulting from the use of generative AI that seem to paraphrase or combine the titles, authors, and/or metadata from one or more real sources. We don't consider a flawed citation to be a hallucitation if it can't plausibly be found online,* or if the title and authors clearly match a real source (even if the rest of the citation is wildly inaccurate). The following table shows the difference between a real citation, a flawed citation, and a hallucitation according to our methodology. The differences are highlighted in red. Real Citation Flawed Citation Hallucinated Citation Maziar Raissi, Paris M. Raissi, P. Maziar Raissi, Paris Perdikaris, and George Perdikaris, and G. Perdikaris, and Em Karniadakis. Karniadakis. George Costanza. A Physics-informed neural Physics-informed neural Deep Learning networks: A deep networks: A deep framework for learning framework for learning framework for physics-informed solving forward and solving froward and neural networks: A inverse problems inverse problems survey on physics involving nonlinear involving nonlinear informed partial differential partial differential reinforcement equations. Journal of equations. Journal of learning. Journal of Computational Physics, Computational Physics, Computational 378:686-707, 2019. doi: 378:686-707, 2019. doi: Physics, 378:686-707, 10.1016/j.jcp. 10.1016%20j.jcp. 2019. doi: 20.1017/ 2018.10.045. 2018.10.045. j.jpc. 2018.12.1942. M. A. Hearst, S. T. M. A. Hearst, S. T. M. A. Hearst (missing Dumais, E. Osuna, J. Dumais, E. Osuna, authors) "Supporting Platt and B. Scholkopf, (missing author) and B. vector machine "Support vector Scholkopf, "Support learning," in IEEE machines," in IEEE vector machines," in Intelligent Systems Intelligent Systems and IEEEE Intelligent and their their Applications, Systems and their Applications, vol. vol. 13, no. 4, pp. Applications, vol. 13w, 11, no. 12, pp. 18-28, July-Aug. 1998, no. 4, (missing page 18-28, July-Aug. 2008 doi: 10.1109/ numbers), July-Aug. , doi: 10.1017/ 5254.708428. 1998, doi: 10.1109/ S0008423924000684 5254.70842. Like GPTZero's AI Detector, Citation Check has an extremely low false negative rate, so we catch 99 out of 100 flawed citations. Because our tool will flag any citation that can't be verified online, the false positive rate is higher. The Future of Peer Review Peer review is an essential part of scholarly publication, yet the current system leaves reviewers and editors outnumbered and outgunned. GPTZero's Citation Check provides two critical benefits to the peer review pipeline. First, using Citation Check together with GPTZero's AI Detector allows users to check for AI-generated text and suspicious citations at the same time, and even use one result to verify the other. Second, Citation Check greatly reduces the time and labor necessary to verify a document's sources by identifying flawed citations for a human to review. We hope that identifying these 50 hallucitations in 50 ICLR submissions shows the value of Citation Check for those facing the submissions avalanche. Our goal is to make the peer review process faster, fairer, and more transparent for everyone involved. Try GPTZero's Citation check for yourself, or reach out to GPTZero's team. If you're interested in being part of our team, we're hiring! Please see our open roles at https://jobs.ashbyhq.com/GPTZero *For example, "Elara Voss, letter to author, October 12, 2024." The Deloitte Citation Situation - $98K Controversy Explained GPTZero used our Citation Check to analyze the 234 page report and identified more than 30 issues out of the total 141 citations, including 19 hallucinations. Using GPTZero's citation check would have saved ~$5000 per citation, all within minutes. [square_log]AI Detection Resources | GPTZeroNazar Shmatko [pexels-ani] Making America Hallucinate Again? GPTZero Detects New Errors in Major Government Report On May 22, the U.S. Presidential Commission to Make America Healthy Again (MAHA), led by health secretary Robert F. Kennedy Jr., released a major report on the causes of chronic diseases in children. Yet within a week, news outlets including NOTUS, the New York Times and Washington Post reported [square_log]AI Detection Resources | GPTZeroPaul Esau [Screenshot] * Investigations * Research * GPTZero Written by Paul Esau Keep reading Investigations Featured Nazar Shmatko, Paul Esau, Alex Cui Investigations Featured Nazar Shmatko, Paul Esau, Alex Cui Deloitte's Citation Situation & GPTZero's Citation Solution GPTZero used our Citation Check to analyze the 234 page report and identified more than 30 issues out of the total 141 citations, including 19 hallucinations. Using GPTZero's citation check would have saved ~$5000 per citation, all within minutes. Nov 26, 2025 * 9 min read Case Study Featured Paul Esau, Alex Cui, Joe Zakielarz Case Study Featured Paul Esau, Alex Cui, Joe Zakielarz Making America Hallucinate Again? GPTZero Detects New Errors in Major Government Report Jul 03, 2025 * 11 min read Share on X Copy link [solid-logo-2] Products * AI Detector * Chrome Extension * Integrations * Plagiarism Checker Resources * Pricing * Sales * Blog * Education Company * About us * Team * Affiliates