Using AI for Automated Reverse Engineering and Reimplementation (zenodo.org)

Researchers announced an AI-assisted reverse engineering framework that reconstructed a bootable prototype of Apple System 7.1 from binaries in just three days—claiming speedups on the order of hundreds of times compared with traditional manual methods. The pipeline orchestrates specialized agents for evidence curation, struct recovery, and code drafting, and couples automated outputs with human review in a tight verification loop. Rather than reporting abstract accuracy metrics, the team validated artifacts end-to-end: screenshots, serial logs, and extracted resources showing Chicago font rendering, menu bar behavior, desktop patterns, and icons to prove functional parity. Technically, the system enforces strict provenance tracking by tying every change back to disassembly bytes or runtime checks under QEMU, enabling reproducibility and auditability of generated code. The approach demonstrates how model-driven decompilation and type/structure recovery can be integrated with deterministic runtime verification to convert reverse engineering from an artisanal process into a systematic workflow for software archaeology and legacy modernization. For the AI/ML community this underscores the value of agent orchestration, artifact-based validation, and provenance-aware generation when applying generative models to safety-critical, correctness-sensitive engineering tasks.