LLM-PF: A LLM-Based Framework for Cryptographic Protocol Flow Extraction via a Novel IR towards Formal Verification

ACL ARR 2026 January Submission6320 Authors

Keywords: Large Language Models, Formal Verification, Cryptographic Protocols, Intermediate Representation, Information Extraction, Protocol Flow

Abstract:

Formal verification is a robust and proven method for ensuring cryptographic protocol security, yet constructing formal models from natural language specifications remains a labor-intensive task demanding specialized expertise.
While Large Language Models excel at text understanding, they struggle to directly generate accurate formal models due to the intricate logic and ambiguities in protocol descriptions.
To bridge this gap, we propose a novel intermediate representation based on finite state machines.
This representation deconstructs protocol logic into modular components, enabling LLMs to extract information step-by-step while ensuring a seamless and complete conversion to formal models.
Building on this design, we developed a comprehensive extraction framework to automate the process.
Extensive evaluations on a diverse set of protocols, ranging from basic to large-scale industrial standards, demonstrate the effectiveness and potential of our approach in automating formal verification.

Paper Type: Long

Research Area: NLP Applications

Research Area Keywords: formal methods with LLMs, information extraction, NLP Application

Contribution Types: NLP engineering experiment, Data resources, Data analysis

Languages Studied: English

Submission Number: 6320