Shanghai Jiao Tong University reverse-engineers European and American industrial software based on the MiniMax large model; all 5,000 core functions will be open-sourced

Recently, "reverse engineer everything" has taken the tech world by storm. This stems from a tool that has garnered hundreds of thousands of stars on GitHub, the world's largest open-source community—REA, or Reverse Engineer Anything. A large number of tech enthusiasts have used it to reverse engineer and heavily modify software and hardware, such as removing startup ads from smart TVs, or setting a smart toilet seat to open from a computer. Some tech enthusiasts have reverse engineered and heavily modified the startup screen of smart TVs. But compared with these small modifications, industrial-grade software is the deep-water zone of "reverse engineering." The reporter learned from the Municipal Economic and Information Technology Commission that Shanghai Jiao Tong University, based on the domestic large model MiniMax, has achieved "re-engineering" of well-known overseas simulation software, image processing and video editing software, among others. Re-engineering refers to first understanding old software and then rewriting it into a better new version, with unchanged functions but an upgraded core. This first large-model-native software reverse-engineering achievement in China was recently released at the Artificial Intelligence Parallel Forum of the 2026 World Laureates Forum (the WLA Forum closed in Lingang, Shanghai, on November 1). China's first large-model-native software reverse-engineering achievement was released at the WLA Forum. The value of software is often built on "source code being unavailable." For example, CAD (computer-aided design), CAE (computer-aided engineering), and CAM (computer-aided manufacturing) software are core tools in the R&D of aviation, shipbuilding, and automotive products, but are basically monopolized by Europe and the United States. The "reverse replication" of such software has always been a professional skill of senior engineers. So-called reverse engineering is when engineers reverse-deconstruct and deduce software without source code, rewrite the code, and reproduce the original software functions. This is like getting a finished dish without a recipe but having to reverse-engineer its ingredients and cooking method. Seven years ago, a group of players spent two years decompiling the action-adventure mobile game Super Mario 64. However, closed-source software is like a "black box," and reverse engineering struggles to "capture the soul," making it impossible to truly restore the core and mechanisms of the software design. The "reverse replication" of Super Mario 64 took two years. Shanghai Jiao Tong University is trying to use "AI + reverse engineering" to overturn this old path. Kong Linghe, a distinguished professor at Shanghai Jiao Tong University and an IEEE Fellow, explained to the reporter that the academic name of this new method is "large-model-based software re-engineering," which relies on large models to truly understand the underlying logic of top overseas industrial software and then redesign a better path to achieve domestic substitution. Specifically, "re-engineering" is divided into three steps. The first step is to understand the "black box's" intentions. This requires modifying the large model itself so that it has native capabilities and can conduct in-depth semantic analysis of the software's core logic, upgrading from "recognizing code" to "being able to dig deep into the intent behind the code." The second step is generative reconstruction. Since it has fully grasped the design purpose of the original software, the large model no longer copies it but rewrites a set of code and optimizes it along the way, such as removing redundancy, improving running speed, and patching security vulnerabilities, to create an optimized and upgraded version. The third step is group cross-quality inspection. Shanghai Jiao Tong University dispatches multiple independent AI agents to conduct "quality inspections" separately—some are responsible for checking logic, some for comparing operational differences between old and new versions, and some for finding vulnerabilities, ensuring that the reconstructed code has no hidden risks. "Re-engineering" is divided into three steps. It is reported that, around the above "three-step" approach, the Shanghai Jiao Tong University team has verified it on multiple domestic and overseas text models. Because the MiniMax text model performs outstandingly in cybersecurity capabilities such as reverse analysis and logic recovery, the "software re-engineering" prototype system ultimately achieved verification and implementation through the MiniMax model. Take the "re-engineering" of a certain US simulation software as an example. This software is widely used in scientific and engineering simulation, and its main body contains 30,000 functions. Kong Linghe's team has completed the "re-engineering" of 5,000 core functions among them, achieving functional alignment and comparable performance. It is reported that Shanghai Jiao Tong University also plans to open-source all 5,000 core functions of the re-engineering, including code, manuals, and test cases. This is of great practical significance for China's layout of AI4S (AI for Science) and dynamic system simulation for major engineering projects. "It goes beyond the pain point that traditional code replication only solves an immediate need, truly grasps the software core, and helps get rid of the risk of being bottlenecked," Kong Linghe said.