AI
First comprehensive review maps AI's growing role in binary code analysis and vulnerability detection
A new systematization of knowledge synthesizes 246 research papers to clarify how machine learning and large language models are reshaping the binary reversing landscape.
Researchers have published the first comprehensive survey of how artificial intelligence is being integrated into binary reversing—the practice of analyzing compiled code to understand software behavior, find security flaws, and investigate malware and firmware.
The Challenge of Reverse Engineering#
Binary reversing has long been central to cybersecurity and software engineering. When source code is compiled into machine language, much semantic information is lost, making it extremely difficult to recover the original logic. This inherent complexity has driven the field to adopt emerging AI technologies—from classical machine learning to large language models (LLMs) and autonomous reasoning systems—to automate and improve analysis.
Organizing a Fragmented Field#
Until now, the growing body of research on AI-augmented binary reversing has remained scattered across different domains, representation formats, and evaluation methodologies. The new systematization, collected from 246 papers published since 2015, organizes the work into 22 distinct binary reversing domains according to the types of inference tasks being performed. The researchers introduce a unified taxonomy that connects:
- Traditional analysis techniques
- How binary code is represented and processed
- Different learning approaches (supervised, unsupervised, reinforcement learning)
- The specific downstream tasks the AI is designed to tackle
- The expanding roles of LLMs and agentic AI systems
What the Review Reveals#
By establishing a common vocabulary and structured framework, the study exposes underlying patterns across seemingly unrelated approaches and identifies persistent gaps in how systems are evaluated. The analysis also highlights technical challenges that remain unresolved and research opportunities where future work could make meaningful progress—pointing toward the next generation of more reliable, evidence-based AI-augmented binary reversing tools suitable for real-world deployment.

