ABOUT

About Reversed with AI

Reversed with AI is a directory of real reverse engineering projects done with AI models such as Claude, Codex and Gemini. Each entry links to the original work and records the model, the tools, the time, the cost and the outcome its author reported.

Why it exists

People now use AI agents to decompile games, decode firmware, document file formats and port old software. The results are spread across blogs, GitHub, X, YouTube and forums, and most posts leave out what it took. This directory puts them in one place with the same facts for each, so you can see what works, with which tools, and how much a person still had to do.

What counts as a project

An AI model did a meaningful part of the reversing: reading disassembly, naming functions, decoding a format or protocol, or writing the port.
The work is public. There is a repo, a writeup, a video or a post we can link to.
The target is fair game under our policy: interoperability, preservation, modding, your own devices, research or defensive analysis.

The full list of what we accept and decline is on What we accept.

How entries are checked

Every entry is read against its primary source: the author's repo, writeup, video or post. We only record what that source says. If the source does not give a time, a cost or how the work was split, the field says Not reported instead of a guess. Figures are reported by the author and are not measured by this site. A daily job checks that the links still work.

Outcome

Works

The result does what the project set out to do. The port runs, the decompiled code builds, the device can be controlled, or the file format can be read and written.

Partial

Real progress with clear gaps. Some functions match, some features work, or the result needs manual fixes to run.

Research

An experiment, benchmark or study of how well AI does at reverse engineering. The point is the findings, not a working tool.

Human effort

A scale from 1 to 5 for how the work was split between the AI and the person, based on what the author describes.

1: Almost all AI
2: Mostly AI
3: Even split
4: Mostly human
5: Human-led

Other fields

Model: The AI model or models the author used, as they named them.
Tools: The harness and reverse engineering tools, for example Claude Code with Ghidra through GhidraMCP.
Time spent: How long the author says the work took. Not reported when the source does not say.
Cost: What the author says they spent on API usage or subscriptions. Not reported when the source does not say.
Where the human stepped in: What the AI got wrong and what the person had to fix or decide.

How sponsors work

The site is paid for by sponsors. Every sponsored placement is labelled and its links use rel=sponsored. Sponsors never affect which projects are listed, their order, or what the guides say. The full rules are on the sponsor page.

Submit or correct an entry

Send a new project through the submit form. If an entry is wrong or out of date, or it is your project and you can add the time, cost or where you stepped in, submit it again with the corrections or message @whyamas on X.

Submit a project @whyamas

Who runs it

Maintained by smyaseen (@whyamas on X), who reviews every submission before it goes live.