TokenAudit: Deterministic Verification of Disclosure-to-Enforcement Consistency in ERC-3643 Tokens with LLM-Based Extraction

• Wyllgner França de Amorim UFPR
• Jáder Louis de S. Gonçalves USP
• Lucas Marques da Cunha UFRO

DOI: https://doi.org/10.5753/webmedia_estendido.2026.33839

Resumo

Real-world asset tokenization embeds compliance rules into the contract itself, yet the natural-language offering document investors read may promise rules the contract does not enforce, a distance we call the disclosure-to-code gap. This paper presents TokenAudit, an open-source tool that audits this consistency for ERC-3643 tokens. A language model extracts a rule checklist from the offering document, the on-chain compliance state is read from the contract, including the modules identified by name, and deterministic rules classify each rule as compliant, divergent, or non-verifiable; the model only structures and explains, it never decides. In an exploratory evaluation over a small set of labeled cases, the deterministic verdict reproduced the expected labels on all evaluated items and the model’s observed error remained confined to extraction, without contaminating the compliance decision.

Palavras-chave: Web3, Large language models, Real-world asset tokenization, ERC-3643, Programmable compliance, Smart contract auditing, Disclosure-to-code gap

Referências

Jianghao Gong, Peiqi Yan, Yue Zhang, Hongli An, and Logan Liu. 2024. Blockchain Meets LLMs: A Living Survey on Bidirectional Integration. arXiv:2411.16809 [cs.CR] [link]

Yanli Jin, Chunpei Li, Peng Fan, Peng Liu, Xianxian Li, Chen Liu, and Wangjie Qiu. 2025. LLM-BSCVM: An LLM-Based Blockchain Smart Contract Vulnerability Management Framework. arXiv:2505.17416 [cs.CR] [link]

Joachim Lebrun, Tony Malghem, Kevin Thizy, Luc Falempin, and Adam Boudjemaa. 2021. ERC-3643: T-REX - Token for Regulated EXchanges. Ethereum Improvement Proposals (EIP-3643). Accessed: 2026-07-27. [link]

Zongwei Li, Xiaoqi Li, Wenkai Li, and Xin Wang. 2026. SCALM: Detecting Bad Practices in Smart Contracts Through LLMs. arXiv:2502.04347 [cs.CL] [link]

Ye Liu, Yue Xue, Daoyuan Wu, Yuqiang Sun, Yi Li, Miaolei Shi, and Yang Liu. 2025. PropertyGPT: LLM-driven Formal Verification of Smart Contracts through Retrieval-Augmented Property Generation. DOI: 10.14722/ndss.2025.241357

Junliang Luo, Xihan Xiong, Zonglun Li, Hong Kang, Xue Liu, William J Knottenbelt, and Katrin Tinn. 2026. SoK of RWA Tokenization: A Systematization of Concepts, Architectures, and Legal Interoperability. arXiv:2604.06608 [q-fin.GN] [link]

Stefan-Claudiu Susan, Andrei Arusoaie, and Dorel Lucanu. 2026. Benchmarking LLM-Based Static Analysis for Secure Smart Contract Development: Reliability, Limitations, and Potential Hybrid Solutions. (2026). arXiv:2605.11163 [cs.CR] [link]

Giorgio Vella, Luca Pennella, and Mark C. Ballandies. 2026. A Taxonomy of Real-World Asset Tokenization for Blockchain-Based Financial Infrastructure. arXiv:2606.08534 [econ.GN] [link]

Olivia Peiyu Wang, Sanna Wong-Toropainen, Daneshvar Amrollahi, Ryan Bai, Tashvi Bansal, Arush Garg, and Leilani H. Gilpin. 2026. Know Your Limits : On the Faithfulness of LLMs as Solvers and Autoformalizers in Legal Reasoning. arXiv:2606.16118 [cs.AI] [link]

Chanuka Wijayakoon, Hai Dong, H. M. N. Dilum Bandara, Zahir Tari, and Anurag Soin. 2025. Legal Compliance Evaluation of Smart Contracts Generated By Large Language Models. arXiv:2506.00943 [cs.SE] [link]

Ning Xia, Xiaolei Zhao, Yimin Yang, Yixuan Li, and Yucong Li. 2025. Exploration on Real World Assets and Tokenization. arXiv:2503.01111 [cs.CR] [link]

Shihao Xia, Mengting He, Shuai Shao, Tingting Yu, Yiying Zhang, Nobuko Yoshida, and Linhai Song. 2026. SymGPT: Auditing Smart Contracts via Combining Symbolic Execution with Large Language Models. 482–512 pages. DOI: 10.1145/3798217

Mounira Nihad Zitouni, Amal Ahmed Anda, Sahil Rajpal, Daniel Amyot, and John Mylopoulos. 2024. Towards the LLM-Based Generation of Formal Specifications from Natural-Language Contracts: Early Experiments with Symboleo. arXiv:2411.15898 [cs.SE] [link]