arXiv subjects

Zhuoran Pan

Publications and source records attributed to Zhuoran Pan.

3 recordsLinked to original sources

arXiv2026

Intent2Tx: Benchmarking LLMs for Translating Natural Language Intents into Ethereum Transactions

The emergence of Large Language Models (LLMs) offers a transformative interface for Web3, yet existing benchmarks fail to capture the complexity of translating high-level user intents into functionally correct, state-dependent on-chain transactions. We present \textsc{Intent2Tx}, a high-fidelity benchmark featuring 29,921 single-step and 1,575 multi-step instances meticulously derived from 300 days of real-world Ethereum mainnet traces. Unlike prior works that rely on synthetic instructions, \textsc{Intent2Tx} grounds natural language intents in real-world protocol interactions across 11 categories, including diverse long-tail Decentralized Finance (DeFi) primitives. To enable rigorous evaluation, we propose an execution-aware framework that transcends surface-level text matching by employing differential state analysis on forked mainnet environments. Our extensive evaluation of 16 state-of-the-art LLMs reveals that while scaling and retrieval-augmentation enhance logical consistency and parameter precision, current models struggle with out-of-distribution generalization and multi-step planning. Crucially, our execution-based analysis demonstrates that syntactically valid outputs often fail to achieve intended state transitions, highlighting a significant gap in current "reasoning-to-execution" capabilities. \textsc{Intent2Tx} serves as a critical foundation for developing autonomous, reliable agents in intent-centric Web3 ecosystems. Code and data: https://anonymous.4open.science/r/Intent2Tx_Bench-97FF .

cs.AI↗

arXiv2026

OMNIINTENT: A Trusted Intent-Centric Framework for User-Friendly Web3

The increasingly complex Web3 ecosystem and decentralized finance (DeFi) landscape demand ever higher levels of technical expertise and financial literacy from participants. The Intent-Centric paradigm in DeFi has thus emerged in response, which allows users to focus on their trading intents rather than the underlying execution details. However, existing approaches, including Typed-intent design and LLM-driven solver, trade off expressiveness, trust, privacy, and composability. We present OMNIINTENT, a language-runtime co-design that reconciles these requirements. OMNIINTENT introduces ICL, a domain-specific Intent-Centric Language for precise yet flexible specification of triggers, actions, and runtime constraints; a Trusted Execution Environment (TEE)-based compiler that compiles intents into signed, state-bound transactions inside an enclave; and an execution optimizer that constructs transaction dependency graphs for safe parallel batch submission and a mempool-aware feasibility checker that predicts execution outcomes. Our full-stack prototype processes diverse DeFi scenarios, achieving 89.6% intent coverage, up to 7.3x throughput speedup via parallel execution, and feasibility-prediction accuracy up to 99.2% with low latency.

cs.CR↗

arXiv2025

Knowledge-to-Jailbreak: Investigating Knowledge-driven Jailbreaking Attacks for Large Language Models

Large language models (LLMs) have been increasingly applied to various domains, which triggers increasing concerns about LLMs' safety on specialized domains, e.g. medicine. Despite prior explorations on general jailbreaking attacks, there are two challenges for applying existing attacks on testing the domain-specific safety of LLMs: (1) Lack of professional knowledge-driven attacks, (2) Insufficient coverage of domain knowledge. To bridge this gap, we propose a new task, knowledge-to-jailbreak, which aims to generate jailbreaking attacks from domain knowledge, requiring both attack effectiveness and knowledge relevance. We collect a large-scale dataset with 12,974 knowledge-jailbreak pairs and fine-tune a large language model as jailbreak-generator, to produce domain knowledge-specific jailbreaks. Experiments on 13 domains and 8 target LLMs demonstrate the effectiveness of jailbreak-generator in generating jailbreaks that are both threatening to the target LLMs and relevant to the given knowledge. We also apply our method to an out-of-domain knowledge base, showing that jailbreak-generator can generate jailbreaks that are comparable in harmfulness to those crafted by human experts. Data and code are available at: https://github.com/THU-KEG/Knowledge-to-Jailbreak/.

cs.CL↗