Financial misinformation in news and analysis can mislead investors, affect market behavior, and weaken trust in financial systems. Detecting this kind of misinformation is particularly difficult when no external evidence or references are available. In many real-world scenarios, users must judge whether a piece of financial text is reliable purely based on the content itself. We build upon RFC-BENCH[1], a newly proposed benchmark for Reference-Free Financial Misinformation Detection that evaluates model reasoning directly from textual content, capturing the contextual complexity of financial news where meaning emerges from dispersed signals.

This shared task invites participants to develop and evaluate systems for detecting reference-free financial misinformation. The goal is to encourage models that can determine whether financial text contains misleading or incorrect claims using only the information contained in the text. The task training dataset provides labeled financial paragraphs and reflects the difficulty of this evaluation setup, in which models must rely on semantic understanding and internal consistency rather than on comparison with external references. We hope this shared task will accelerate progress toward more reliable and context-aware misinformation detection in financial language processing.

[task prompt] denotes the instruction for the task (e.g. Please determine whether the news is True or False). 

This shared task uses the RFC-Bench[1] dataset (Reference-Free Counterfactual Financial Misinformation Benchmark), which was introduced to evaluate models on reference-free financial misinformation detection. The dataset contains paragraphs extracted from real financial news, each labeled as containing misleading or incorrect financial information. Unlike setups that require external evidence or background knowledge, this dataset is designed so that models must classify misinformation solely based on the content of the paragraph.

Dataset Contents

Each instance in the dataset consists of the following:

• paragraph: A standalone piece of financial news text.

• label: A binary indicator showing whether the paragraph is real news or perturbed news (True or False).

This reference-free structure reflects realistic scenarios where no supporting evidence is provided, and models must rely on semantic reasoning and content coherence to detect misinformation. The dataset highlights the limitations of methods that depend on external context and encourages development of more robust and context-aware detection systems.

Training Data Examples:

We evaluate misinformation detection as a binary classification task using Accuracy, Precision, Recall, and F1 score. Accuracy measures the proportion of correctly classified instances over all predictions. Recall quantifies the fraction of true positive cases successfully identified, while Precision measures how many predicted positives are actually correct. The F1 score is defined as the harmonic mean of Precision and Recall, balancing false positives and false negatives into a single metric.

To measure the risk of data leakage from the test set used in the training of a model, the Model Cheating, participants need to upload their final model to hugging face and the necessary scripts for Model Cheating Detection. 

Please choose a unique team name and ensure that all team members provide their full names, emails, institutions, and the team name. Every team member should register using the same team name. We encourage you to use your institution's email to register.

• The up-to-date AAAI 2026 MUST be used for your submission(s). Accepted papers proceedings will be published at ICWSM workshops. The main text is limited to 4 pages. The appendix is unlimited and placed after references.

• The paper title format is fixed: "[Model or Team Name] at the Financial Misinformation Detection Challenge Task: [Title]".

• The reviewing process will be single-blind.

• Shared task participants will be asked to review other teams' papers during the review period. 

• Submissions must be in electronic form using the paper submission software linked above.

• At least one author of each accepted paper should register and present their work in person at MisD@ICWSM2026 workshop. Papers with “No Show” may be redacted. Authors will be required to agree to this requirement at the time of submission.

• Yuechen Jiang - University of Manchester, UK

• Yuyan Wang - University of Manchester, UK

• Peter Carragher - Carnegie Mellon University, US

• Tianlei Zhu - University of Columbia, US

• Zhiwei Liu - University of Manchester, UK

• Yupeng Cao - Stevens Institute of Technology, US

• Jimin Huang - University of Manchester, UK; The FinAI, US

• Sophia Ananiadou - University of Manchester, UK; Archimedes RC, Greece

• Contact email: fmd.finnlp@gmail.com