AI Reverse Engineering Shakes Software Industry

AI tools replicate Adobe, Microsoft software, sparking legal and market concerns

Brandon Thomas, a U.S. software developer, recently created a software called “ArtCraft” using artificial intelligence (AI) that offers similar functionalities to seven programs, including Adobe’s Photoshop, Illustrator, and Premiere Pro. He later developed programs mimicking Microsoft’s Word, Excel, and PowerPoint.

Thomas, who founded the AI video and voice generation company “Storyteller.ai,” released these programs for free. As news spread that core features of existing commercial software could be replicated with AI, financial markets reacted immediately. HSBC, a British financial firm, lowered its investment rating for Adobe from “Buy” to “Hold” on October 13 and reduced its target stock price from $310 to $267, a 13.9% decrease.

The replication of Adobe and Microsoft software has highlighted AI-driven “reverse engineering” as a new threat to the software industry. Reverse engineering involves analyzing the internal structure and operation of completed software to recreate similar functions with new code.

It is akin to disassembling a newly released car to understand its core technologies, such as the engine and transmission, and then building a similar vehicle. With AI supporting analysis and development, experts predict that the time and cost required to replicate existing software functions could decrease significantly.

AI-Driven ‘Software Reverse Engineering’

Reverse engineering technology for software has existed before. The U.S. National Security Agency (NSA) developed a tool called “Ghidra” in the late 1990s to estimate original code from executable files for analyzing adversarial malware. This tool allowed analysts to understand the structure and operation of malware or infected software.

Ghidra, which was hidden for nearly 20 years, was publicly released in March 2019 and later transitioned to open-source, sparking significant interest in software reverse engineering. However, the code restored by Ghidra lacked all comments added by programmers, making it extremely complex. With hundreds of thousands of lines of code, even multiple software experts required months to analyze it.

AI has overcome these challenges. The recent public release of the AI reverse engineering tool “REA (Reverse Engineer Anything)” on GitHub, the world’s largest software development platform, has accelerated its spread. REA connects AI coding tools like OpenAI’s Codex and Anthropic’s Claude Code with specialized reverse engineering tools to analyze a program’s structure and operation.

Notably, REA allows users to input only necessary information from vast analysis results, reducing token (data processing unit) usage and costs. This lowered barrier to analyzing existing programs could lead to more attempts to recreate software functions using AI.

Document Editing Features Recreated in 3 Minutes

A reporter tested REA and OpenAI Codex to recreate parts of “Novel,” an open-source code-based document editor. Investigating the program’s structure with REA took approximately 7 seconds. Codex then read the open-source code and implemented and verified editing features like titles, bold text, and bullet points in 3 minutes and 3 seconds. All 15 tests comparing the original and the result matched. After spending an additional 3 minutes and 44 seconds, the editing interface was created, passing 14 operational tests. When the reporter input Korean text and applied titles and bold formatting, it functioned normally.

AI reverse engineering is gaining attention because it can analyze not only the functions but also the design and development know-how accumulated by software companies over decades. Unlike “vibe coding,” where AI creates programs based on described functions, reverse engineering analyzes completed programs to understand their implementation methods.

Reactions in the software industry are mixed. Pieter Levels, a Dutch developer and entrepreneur, stated, “Software is dead. People will create programs they need or ask AI to do so whenever required.” He suggests that as AI enables users to create necessary programs themselves, fewer people will pay expensive subscription fees for commercial software and instead implement only required functions.

Conversely, Brazilian developer Fábio Akita refuted the “end of the software industry” theory, stating, “AI-generated alternatives have errors and missing features, and it is difficult to analyze all functions of commercial products externally.”

Domestic Attempts to Recreate ‘Hangul’

In South Korea, a developer publicly revealed “Tangul 2026,” a program that can read and edit HWP and HWPX files, the document formats of Hancom’s “Hangul.” The developer explained that the program was created for educational and demonstration purposes. While it does not fully replicate all features of Hangul or MS Word, it marks the emergence of domestic attempts to recreate existing commercial software functions using AI.

As AI-driven software reverse engineering and function replication attempts increase domestically and internationally, legal disputes over copyright infringement are expected to rise. South Korea’s Copyright Act permits code analysis under certain conditions, such as investigating a program’s operational principles or ensuring compatibility with other programs. However, using the obtained information beyond permitted purposes or unauthorized replication of protected code can lead to copyright infringement issues.

A tech industry insider stated, “AI allows companies to implement functions that took decades to develop in a much shorter time. Conflicts over copyright and technology protection could intensify between companies that invested heavily in R&D and latecomers that replicate features cheaply.”

This article has been translated by Upstage Solar AI.