decomp.dev turns off its public index, citing AI decompilations

On 15 October 2026, decomp.dev replaced its front page with a note titled “Sunsetting decomp.dev”, signed @encounter. The note says the site will keep tracking existing projects and will keep serving individual project pages and historical data, and that the main index and adding new projects are disabled.

404 Media, Jason Koebler, carries a publish time of 16:31 UTC the same day. The byline on the page reads 12:31 PM. Koebler reports that the site is shutting down because AI made matching decompilation too easy. The site’s own note is narrower than that headline: the public index and the new-project form stop, and the project pages stay.

What matching decompilation means

Decompedia’s intro describes a matching decomp as a functionally identical reconstruction of a game’s source from the machine code it shipped with, written so that the game’s original compiler produces a byte-for-byte copy of the original. Encounter’s note uses that standard as the thing the site was built to track. He writes that matching decompilation “has been a brutally difficult, time consuming task, on the scale of years for any moderately-sized game with a number of active contributors,” and that decomp.dev “was originally designed to gamify decompilation: watching the progress go up was a fun way to stay motivated when every fraction of a percent took real human effort.”

What the note says changed

The note says AI experiments in decompilation went on for years with mixed results. It says that with Opus 4.6 and GPT-5.4 in early 2026, the models appeared to know the community’s tools and could get decent results in a clean, documented codebase, still with a lot of human cleanup, and at a cost the note calls prohibitive. It then names recent models, “like GPT-6-Astra”, as better at long instructions, at looking up symbols or debug information, at writing natural code, and at trying many source forms until one matches. Those model names are the note’s. This story does not add others.

Encounter writes: “It is now clear that, with the right guidance, human input, and code reviews, AI can expedite the decompilation process in a significant way. What is also now clear is that the community aspect of decompilation has been broken.”

He lists three practices. One is forking an open project to “finish” it with AI, including projects that reject AI for legal or ethical reasons, or that accept AI and still refuse drive-by changes. He calls these “hostile forks” that “put little to no effort into meaningfully contributing to existing projects.” Another is racing to publish a port from a decompilation the porter did not contribute to. He writes that “anyone with an AI subscription can manage to create a port or static recompilation that functions at a basic level,” and that “opportunists have frequently raced to release a half-baked, barely-functional port.” The third is using the word decomp to mean a PC port. He says the progress board has “warped decompilation into a hollow numbers game, exacerbated by folks who just want to burn tokens to make number go up.”

He also writes that he will not ban AI-assisted projects, because he thinks AI can help when it is reviewed, and that as a sole hobbyist he will not define and police “good” decompilation across the projects on the site. The note says the site’s own code was “entirely human written,” was never under an open source license, and may not be copied into a replica.

Koebler’s article quotes those passages and frames the decision next to AI-made clones of commercial software and a copyright argument. Encounter’s mention of legal or ethical objections is his report of why some projects refuse AI. This article does not judge whether any decompilation, fork, or port infringes a copyright.

The operational change on the page is the one in the note: the index and new submissions are off, existing project pages and historical data remain, and the reason given is the admin’s.