Open-source project REA goes viral: AI lowers the barrier to reverse engineering, games and software are being cracked at scale

With Double 11 here, viewers who love console gaming have surely noticed that the previously unloved Japanese version of the PS5 has suddenly become hot property—the good-deal post published today (yesterday) already has over 100 comments underneath.

How did a region-locked console that can only play Japanese games suddenly become so popular? The answer is in those 100-plus comments.

That's right—the PS5 has been cracked. Not only can it run games for free, it can even run downloaded models to do local AI inference. A $2,500 console used as a ChatGPT machine—Jensen Huang would want to hit someone after seeing that.

In fact, it's not just the PS5—the entire software world is being cracked.

Various games are being stitched together—Spider-Man braving Gotham City, My World War GTA Police.

Mario vs. Elden Ring

Some people, after reverse-engineering games, have even turned them directly into web versions. When a guy abroad opened his browser and saw it was Call of Duty, his jaw dropped. If we'd had this back then, who would have bothered secretly copying games onto a USB drive to play in computer class!

The cracking scene worldwide is in a thriving trend, forming a sharp contrast with everyone's stock accounts.

All one can say is that the big wave—thin muscle is already behind the times; big cracking is the real hotspot right now.

REA, reverse-engineer everything!

All of this is inseparable from a project on GitHub—REA. Reverse Engineering Anything, which translates to "reverse-engineer anything."

Although it went live on GitHub in July, it only got widespread attention around October 7. In five days, its GitHub stars soared from under 10,000 to over 140,000, and now it's shot past 300,000.

What is reverse engineering? The most direct way to understand it: say we want to build a car—first comes the idea, then the design drawings, then the car is produced. That's forward engineering.

Reverse engineering is: we first buy a car, tear it apart, measure the specs of every component, then work backward to the drawings and produce it. In plain terms, it's counterfeiting.

So what is there to reverse-engineer in software? Can't I just Ctrl C + Ctrl V and copy it?

Yes, you can. But most software doesn't provide source code. Ctrl C + Ctrl V can only copy that 01001 machine code, which humans can't understand, so no modifications can be made. If you want to make a small change, or crack it, what do you do?

You can only do software reverse engineering—from 01001 back to assembly language, and then back to high-level languages like Python or C.

Because it's so complex, reverse engineering isn't work a human can do. An experienced reverse engineer analyzing a program of medium complexity might have to grind away for months on end.

What REA does is teach AI this craft overnight.

REA itself is not an AI, nor an Agent. More accurately, it's a collection of MCP services plus a skill pack (Skill). You can think of it as a "reverse-engineering textbook + toolbox"—it doesn't do the work itself, but it can teach AI to do the work.

With REA's backing, cracking and reverse engineering have suddenly become something anyone can pick up.

REA delivers, everyone benefits

The first to benefit are the vast number of gamers.

Old mods were just reskins—swapping a Call of Duty gun into GTA looks intimidating, but the damage is still the original value.

Now?

Blocky Steve wielding a diamond sword faces off against the police in Los Santos; TNT explosions and arrow trajectories all use Minecraft's physics engine. The two games achieve data interoperability at the engine level—place a block wall in Minecraft and a collision body is automatically generated in GTA, and cars really do crash into it and bounce off.

Even more extreme, someone stitched Tarkov, Skyrim, and Spider-Man into one game—Spider-Man swings on webs through Skyrim Province while holding a Tarkov assault rifle, firing while swinging. What genre is that? Open-world action shooter swinging RPG?

And that's just games. Remember what the A in REA stands for? Anything!

Operating systems inevitably got reverse-engineered too—first Hackintosh.

For a long time, the biggest obstacle to installing macOS on an ordinary PC was the graphics card. Since 2019, Apple has kicked NVIDIA drivers out of the system, so if you want performance you can only use AMD cards.

Then recently a guy managed to produce an unofficial driver that lets the entire RTX 20/30/40/50 series run on macOS.

Then there's the PS5. The PS5 itself runs a customized Unix system. The cracking team found a memory vulnerability by reverse-engineering the WebKit browser, then reverse-engineered the kernel to gain privilege escalation, and took control of the whole machine.

With root access, players went wild—someone ported the AI inference engine llama.cpp onto it and chatted with AI by typing on the controller.

Even more absurd is the PSBox project: someone reverse-engineered Microsoft Xbox cloud gaming's streaming protocol, letting the PS5 run Xbox Cloud Gaming at 1080p 60fps. Sony's hardware running Microsoft's service playing Microsoft's games—NTR truly achieved.

Since all this can be done, running iOS on an Android phone is naturally no problem either, right?

There's also reverse-engineering IoT protocols—someone used AR glasses to create a very sci-fi effect for adjusting the air conditioner temperature. And someone else directly controls their home toilet from a Mac.

The real engine: not REA, but cheapness

A year ago, any one of these things would have been enough for the entire tech circle to discuss for a month. Now they're all clustered together in September and October—why?

After carefully reading through each reverse-engineering project, I found that REA can only be considered the fuse that drew public attention; for people who truly care about reverse engineering, the models updated over the past six months are the real booster.

Fable 5.1 and GPT-6 Astra brought a qualitative change to AI's code comprehension—not just "writing" code, but a major enhancement in the ability to "read" code. AI can finally understand decompiled pseudocode that even humans find headache-inducing. But these two models have a fatal problem: they're expensive. $10 input, $50 output—running one project can easily burn thousands of dollars.

The turning point was the release of Claude Opus 5.5: its performance is close to Fable 5.1, but the cost dropped to one-third of before. You can push it hard without feeling the pain.

So rather than saying REA made us realize the arrival of the great cracking era, cheap and easy-to-use large models like Opus 5.5 are the real engine of the great cracking era.

Facing the great cracking era, if you're just playing games and running some free PS5 games, that's boring.

More than games

A cryptocurrency trader used REA to reverse-engineer TradingView's paid features.

According to his account, every year he had to pay several hundred dollars for a membership to a piece of software called TradingView. What does this software do? After a quick look, it's roughly the crypto-market equivalent of Tonghuashun.

Its main function is to customize K-line charts in various ways, arrange and combine technical indicators like RSI and MACD, and even DIY them however you want. To experience it, you need to pay a membership fee of tens of dollars every month.

This guy used REA to have AI reverse-engineer how the chart refreshes when switching stock tickers, how the UI responds when overlaying indicators, and how data links up when drawing trend lines. He fed it to Claude Opus 5.5 and rebuilt from scratch a free charting engine with similar functionality. He used it to build himself a 24-hour trading system.

Not only does he no longer pay membership fees, he also built a quantitative system. Looking at his latest tweets, he has now set his sights on the Bloomberg terminal.

But what really makes the entire software industry uneasy isn't financial tools—it's the ArtCraft project. Brandon Thomas, an engineer in Atlanta, got fed up with Adobe's renewals.

Using Opus 5.5 plus Rust, he cranked out seven open-source alternatives in one go. PhotoCraft targets Photoshop, FilmCraft targets Premiere, Lightcraft targets Lightroom, and so on. The PS alternative hit 70,000 stars on GitHub. The Adobe suite costs $779 a year; ArtCraft is free.

What's most intriguing is his method: not reverse engineering, but an approach called "clean room design."

What is clean room design? You form two teams that are completely isolated from each other. One team is responsible for writing every feature and every input/output behavior into a purely textual technical specification.

The other team takes the specification written by the first team and writes a new piece of software from scratch with exactly the same functionality.

Looking at code to produce a feature list is legal, right? I just want to learn what problems your software has, right?

Writing a program based on the feature list is obviously legal too, right?

In other words, this completely avoids copyright issues—even if Adobe wanted to sue, it couldn't.

The most famous example of clean room design is Compaq cloning the IBM PC BIOS in 1982. At the time, IBM launched the first personal computer, the IBM PC. Anyone could buy the hardware components, but the BIOS was written by IBM and protected by copyright. Without a compatible BIOS, your computer couldn't run the IBM PC software ecosystem.

So Compaq adopted the design approach described above, and ultimately Compaq won its lawsuit against IBM. The court ruled that this approach did not constitute infringement because the new code was independently created. This precedent directly gave rise to the entire IBM-compatible machine industry—the predecessors of all later PCs.

Of course, ArtCraft is currently only 25% to 45% complete, and there's still a gap before it can truly replace Adobe. But the thing is, AI spits out several versions a week, and this catch-up speed is something traditional software development wouldn't dare imagine.

In the era of universal cracking, what's left?

At the time of writing this article, the pace of the great cracking shows no sign of slowing. On Twitter, someone just reverse-engineered the algorithm of a major brand's fitness band; someone else removed TV startup ads. This vibrant scene of all things competing to flourish feels like returning to the wild early days of Internet 1.0—everything free, all for one and one for all.

For software companies, are they about to close their doors and settle accounts? Not necessarily.

A guy on Bilibili ran a test: software with encryption and packing protection is still something AI can't crack. So future software may all need encryption systems—like how bicycles didn't used to need locks, but now need two.

But the question that really keeps people up at night is: if the release of one new model can spawn so many reverse-engineering breakthroughs, what happens when even stronger models come out and can easily handle encryption and unpacking too?

English text is being prepared.