May 13, 2026
WinRAR? More like Win-Drama
Rars: a Rust RAR implementation, mostly written by LLMs
AI built a RAR tool for £40, and the comments instantly turned into a trust war
TLDR: A developer says AI helped build a free RAR file tool in weeks instead of years, but readers are fighting over whether it’s brilliant, legally risky, or impossible to trust. The biggest debate isn’t the software itself—it’s whether machine-made code can be believed.
A programmer says he used chatbots plus a lot of yelling to build a free tool that can create old-school RAR archive files, the compressed folders many people remember from the WinRAR era. The wild part? He claims a job that once felt like a five-year mountain climb got done in about five weeks of evenings, for roughly £40 in tokens. That alone was enough to summon the crowd, but the real fireworks started when readers asked the question hanging over the whole project: can anyone actually trust code mostly written by machines?
The comment section quickly split into camps. One side was impressed: this thing is sloppy, slow, and bigger than it should be, but it works, and that’s a huge deal for a file format that never had a truly free version. One commenter basically gave it the classic engineer blessing: optimization can come later, you already survived the impossible part. Another showed up as the repo’s third starrer, which is exactly the kind of scrappy underdog energy the internet loves.
But the skeptics came in hot. One asked, in effect, who even owns machine-written code, and whether this project could be legally messy. Another cut straight to the nerve: how do we know it’s actually correct? Then there was the delicious side drama about the tool almost earning an OpenAI ban after an AI wandered into piracy-adjacent territory. That tidbit had readers sniffing around for a cybersecurity scandal like it was celebrity gossip. In the end, the vibe was equal parts admiration, suspicion, and popcorn-worthy chaos.
Key Points
- •The author says Rars, a Rust implementation of the RAR format, was built in about five weeks with substantial help from OpenAI Codex 5.5 and Claude Opus 4.7.
- •The project required reconstructing RAR specifications because formal specs were lacking and the available unrar source was not fully free software.
- •To document the format, the author used existing decompressor sources, iterative LLM-assisted documentation, and reverse-engineering of DOS and Windows binaries with Ghidra and DOSBox-X.
- •The resulting codebase is described as roughly 55,000 lines, functional, and able to support every version of the RAR file format, though not optimized for speed.
- •The author reports that AI-generated documentation about WinRAR registration bypass triggered OpenAI cybersecurity flags, leading to verification requirements and removal of that material from the project history.