Autonomous Fact-Checking: Integrating Local Inference and Adversarial NLP Attacks
Resumo
Digital misinformation threatens democratic discourse, especially in low-resource languages like Brazilian Portuguese. We present a fact-checking pipeline comparing structured claim decomposition (Claimify) against full-text extraction across four retrieval-augmented classifiers: gemini-2.5-flash-lite and three local open-weight models (Llama-3.1-8B, Qwen2.5-7B, Gemma-2-9B). Robustness is assessed under characterlevel and synonym perturbations. On a pilot sample, the API-based model shows a preliminary macro-F1 advantage for decomposition, which the three local models fail to replicate consistently. We also observe a systematic bias toward the false label across models.
Referências
Campos, R., Sequeira, R., Nerea, S., Cantante, I., Folques, D., Cunha, L. F., Canavilhas, J., Branco, A., Jorge, A., Nunes, S., Guimarães, N., and Silvano, P. (2026). ClaimPT: A portuguese dataset of annotated claims in news articles. In Advances in Information Retrieval (ECIR 2026), volume 16486 of Lecture Notes in Computer Science, pages 544–560. Springer.
Garcia, D., Silva, A., and Santana, C. (2022). Fake news detection in brazilian portuguese: Datasets and current challenges. In Proceedings of the Brazilian Symposium on Information Systems, pages 115–122.
Guo, Z., Schlichtkrull, M., and Vlachos, A. (2022). From ready-to-wear to ready-to-match: A survey on automated fact-checking. ACM Computing Surveys, 55(6):1–34.
Hu, Q., Long, Q., and Wang, W. (2025). Decomposition dilemmas: Does claim decomposition boost or burden fact-checking performance? In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 6313–6336.
Jin, D., Jin, Z., Zhou, J. T., and Szolovits, P. (2020). Is BERT really robust? a strong baseline for natural language attack on text classification and entailment. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 8018–8025.
Lewis, P., Perez, E., Piktus, A., Petroni, F., Lewis, V., Riedel, S., and Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. In Advances in Neural Information Processing Systems, pages 9459–9474.
Liu, F., Abuadbba, A., Moore, K., Nepal, S., Paris, C., Wu, J., Yang, J., and Sheng, Q. Z. (2025). Adversarial attacks against automated fact-checking: A survey. arXiv preprint arXiv:2509.08463.
Metropolitansky, I., Goldstein, A., and Shahaf, D. (2025). Claimify: Decomposing documents into atomic verifiable claims using large language models. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 412–426.
Morris, J., Lifland, E., Yoo, J. Y., Grigsby, J., Jin, D., and Qi, Y. (2020). TextAttack: A framework for adversarial attacks, data augmentation, and adversarial training in NLP. In Proceedings of the Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 119–126.
Pan, J., Wang, X., and Gao, Y. (2023). A survey on retrieval-augmented text generation for large language models. arXiv preprint arXiv:2304.09400.
Schlichtkrull, M., Guo, Z., and Vlachos, A. (2023). AVeriTeC: A dataset for real-world fact checking with evidence. arXiv preprint arXiv:2305.13117.
Silva, J., Santos, R., Almeida, T., and Pardo, T. (2020). Fake.Br corpus: A corpus for fake news detection in brazilian portuguese. In Proceedings of the International Conference on Computational Processing of Portuguese, pages 416–425.
Silva, R. M., Amamou, H., Ferraz, L. B. S., da Silva, F. K. A., and Avila, A. R. (2025). Fake news detection in portuguese under large language model-generated content. Journal of the Brazilian Computer Society, 31(1):1149–1166.
Thorne, J., Vlachos, A., Christodoulopoulos, C., and Mittal, A. (2018). FEVER: a large-scale dataset for fact extraction and VERification. In Proceedings of the Human Language Technologies: NAACL, pages 809–824.
Wang, W. Y. (2017). Liar, liar pants on fire: A new benchmark dataset for fake news detection. In Proceedings of the Annual Meeting of the Association for Computational Linguistics, pages 422–426.
Zuccon, G., Koopman, B., and Palotti, J. (2022). The landscape of fact-checking resources for low-resource languages. Journal of Artificial Intelligence Research, 74:1435–1461.
Garcia, D., Silva, A., and Santana, C. (2022). Fake news detection in brazilian portuguese: Datasets and current challenges. In Proceedings of the Brazilian Symposium on Information Systems, pages 115–122.
Guo, Z., Schlichtkrull, M., and Vlachos, A. (2022). From ready-to-wear to ready-to-match: A survey on automated fact-checking. ACM Computing Surveys, 55(6):1–34.
Hu, Q., Long, Q., and Wang, W. (2025). Decomposition dilemmas: Does claim decomposition boost or burden fact-checking performance? In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 6313–6336.
Jin, D., Jin, Z., Zhou, J. T., and Szolovits, P. (2020). Is BERT really robust? a strong baseline for natural language attack on text classification and entailment. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 8018–8025.
Lewis, P., Perez, E., Piktus, A., Petroni, F., Lewis, V., Riedel, S., and Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. In Advances in Neural Information Processing Systems, pages 9459–9474.
Liu, F., Abuadbba, A., Moore, K., Nepal, S., Paris, C., Wu, J., Yang, J., and Sheng, Q. Z. (2025). Adversarial attacks against automated fact-checking: A survey. arXiv preprint arXiv:2509.08463.
Metropolitansky, I., Goldstein, A., and Shahaf, D. (2025). Claimify: Decomposing documents into atomic verifiable claims using large language models. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 412–426.
Morris, J., Lifland, E., Yoo, J. Y., Grigsby, J., Jin, D., and Qi, Y. (2020). TextAttack: A framework for adversarial attacks, data augmentation, and adversarial training in NLP. In Proceedings of the Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 119–126.
Pan, J., Wang, X., and Gao, Y. (2023). A survey on retrieval-augmented text generation for large language models. arXiv preprint arXiv:2304.09400.
Schlichtkrull, M., Guo, Z., and Vlachos, A. (2023). AVeriTeC: A dataset for real-world fact checking with evidence. arXiv preprint arXiv:2305.13117.
Silva, J., Santos, R., Almeida, T., and Pardo, T. (2020). Fake.Br corpus: A corpus for fake news detection in brazilian portuguese. In Proceedings of the International Conference on Computational Processing of Portuguese, pages 416–425.
Silva, R. M., Amamou, H., Ferraz, L. B. S., da Silva, F. K. A., and Avila, A. R. (2025). Fake news detection in portuguese under large language model-generated content. Journal of the Brazilian Computer Society, 31(1):1149–1166.
Thorne, J., Vlachos, A., Christodoulopoulos, C., and Mittal, A. (2018). FEVER: a large-scale dataset for fact extraction and VERification. In Proceedings of the Human Language Technologies: NAACL, pages 809–824.
Wang, W. Y. (2017). Liar, liar pants on fire: A new benchmark dataset for fake news detection. In Proceedings of the Annual Meeting of the Association for Computational Linguistics, pages 422–426.
Zuccon, G., Koopman, B., and Palotti, J. (2022). The landscape of fact-checking resources for low-resource languages. Journal of Artificial Intelligence Research, 74:1435–1461.
Publicado
19/10/2026
Como Citar
DUARTE, Rômulo Henrique Nascimento; LIRA, Humberto Nunes de; PEREIRA, Vivianny Khatly Medeiros; ARAÚJO, Isaac Sebastian Lima de; VASCONCELOS, Allan Gabriel da Cunha; BARBOSA, Yuri de Almeida Malheiros.
Autonomous Fact-Checking: Integrating Local Inference and Adversarial NLP Attacks. In: SIMPÓSIO BRASILEIRO DE TECNOLOGIA DA INFORMAÇÃO E DA LINGUAGEM HUMANA (STIL), 17. , 2026, Cuiabá/MT.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 113-125.
DOI: https://doi.org/10.5753/stil.2026.26643.
