project · DiggingBeagle record
ContractLens (malicious smart-contract research)
A 2026 academic framework for identifying malicious-contract-discriminative activations in frozen language models and using them to screen labeled EVM smart contracts.
- Entity kind
- project
Record description
ContractLens is the model-interpretability and malicious-contract classification research framework described by Shuyi Miao, Xinyi Huang, Wangjie Qiu, Fei Shen, Jinchun He, Zhiming Zheng, and Tat-Seng Chua in an October 4, 2026 preprint. The work draws on Beihang University, Beijing University of Posts and Telecommunications, and the National University of Singapore. It locates useful layers and MLP neurons through linear probes, counterfactual malicious-logic removal and code-surface perturbation controls, then trains a lightweight classifier on selected activations without changing the backbone model. It is a research method, not a verified deployed monitoring service, and is distinct from unrelated legal contract analyzers and hackathon smart-contract products using the same name.