How REA Cracked Closed-Source Software Wide Open — Transcript
Full transcript
- 0:00Two weeks ago, a free repository called
- 0:02REA, short for reverse engineer
- 0:05anything, quietly went nuclear on
- 0:08GitHub, exploding past 115,000
- 0:12stars and 25,000 forks in a matter of
- 0:15days. With a single terminal command, it
- 0:18connects AI coding agents directly into
- 0:21the US National Security Agency's binary
- 0:23decompiler, allowing anyone to take a
- 0:26locked, [music]
- 0:26closed-source .exe app or video game and
- 0:30strip-mine it back into clean C++ source
- 0:33code in less than 48 hours. Developers
- 0:36are already using it to mass decompile
- 0:38commercial games and crack proprietary
- 0:40hardware. But when independent
- 0:42researchers ran 482 controlled tests on
- 0:46this tool just days ago, they uncovered
- 0:48a twist that proves closed-source
- 0:50software isn't just vulnerable, it might
- 0:53already be dead.
- 0:58>> [snorts]
- 0:59>> Up until this month, if you wanted to
- 1:01use AI to clone a piece of paid
- 1:03software, you had to vibe code it.
- 1:07You showed an AI a screenshot of
- 1:09Photoshop or a video game, and the AI
- 1:11tried to guess how the code worked from
- 1:13the outside.
- 1:14For simple apps, that worked. But the
- 1:17moment an application had complex game
- 1:19physics, custom audio engines, or
- 1:22encrypted network checks, the AI hit a
- 1:24brick wall. Because it couldn't see the
- 1:27actual source [music] code locked inside
- 1:29the .exe file.
- 1:31Normally, when a studio finishes writing
- 1:33a program, they run it through a
- 1:35compiler that strips away every folder,
- 1:38every variable name, and every human
- 1:40word, crushing the code down into
- 1:43millions of lines of raw machine
- 1:45assembly.
- 1:46Even if you drop that file into elite
- 1:48forensic tools like IDA Pro or the NSA's
- 1:51Ghidra decompiler, all you got back was
- 1:54a scrambled maze of anonymous memory
- 1:56addresses that took human hackers years
- 1:58to decipher by hand.
- 2:01REA bypassed that human bottleneck
- 2:03overnight.
- 2:04Built by a developer named Moluto, the
- 2:07tool uses the model context protocol, or
- 2:10MCP, to give AI agents like Claude code,
- 2:13OpenAI Codex, and Cursor direct hands-on
- 2:17control over Ghidra, IDA Pro, Binwalk,
- 2:20and live network interceptors.
- 2:23Instead of a human staring at nameless
- 2:25assembly code for 6 months, the AI agent
- 2:28opens the binary inside Ghidra
- 2:30headlessly, traces thousands of function
- 2:33calls across the memory graph in
- 2:34seconds, and rewrites the unreadable
- 2:37assembly back into clean C++ or [music]
- 2:39Rust.
- 2:41Even crazier, it doesn't guess.
- 2:44In a test on the classic PC game DX
- 2:46Ball, the agent extracted a compiled
- 2:48[music] x86 audio routine, rebuilt it in
- 2:52clean C, tested it against 3,205
- 2:55execution states, and matched 63 out of
- 2:5963 original compiled bytes with 100%
- 3:02accuracy.
- 3:04And within hours of that demo going
- 3:06live, the gaming underground turned it
- 3:08loose on commercial games.
- 3:14Wait until you see what happened when
- 3:16game modders got their hands on this
- 3:18workflow.
- 3:20Historically, decompiling a single video
- 3:22game back into native source code was
- 3:25one of the hardest feats in computer
- 3:27science. When fans manually decompiled
- 3:30Super Mario 64 and Zelda: Ocarina of
- 3:33Time, it took teams of programmers years
- 3:36of agonizing work. Therefore, only a
- 3:39handful of legendary games were ever
- 3:41decompiled. Two weeks ago, that
- 3:44multi-year barrier collapsed into a
- 3:46single [music] weekend.
- 3:48Look at a viral project called Isaac
- 3:50Reborn. The original 2011 version of The
- 3:53Binding of Isaac was built inside dead
- 3:56Adobe Flash code, permanently trapped at
- 3:59a laggy 30 frames per second and a
- 4:01blurry low resolution. A developer
- 4:04pointed an AI decompilation pipeline
- 4:06straight at the game's binary and
- 4:08automatically translated the entire
- 4:11codebase into native C++ in just days.
- 4:15Right out of the box, a 15-year-old
- 4:17Flash game was running natively on
- 4:19modern PCs at 1440p and a locked 165
- 4:24frames per second.
- 4:25And Isaac Recomp was just the opening
- 4:28shot. Within days, Hacker News and
- 4:30GitHub were flooded with AI decompiled
- 4:33game engines. One developer used AI and
- 4:36Ghidra to reconstruct the proprietary
- 4:39Refractor 2 engine from Battlefield 2
- 4:41into modern C++ 20. Another extracted a
- 4:45full 3D engine from Subway Surfers.
- 4:48Another translated the cult classic
- 4:50shooter Gun Z the Duel straight into
- 4:53memory-safe Rust. [music] While a new
- 4:55experimental project called NPS5
- 4:58launched with a wild goal, using
- 5:01automated binary translation to convert
- 5:03PlayStation console executables [music]
- 5:05straight into native PC ports.
- 5:08For gamers, watching abandoned classics
- 5:10get rescued from dead hardware looked
- 5:12like a golden age. But within days,
- 5:15developers realized this pipeline
- 5:17doesn't just work on old video games.
- 5:19[music] It works on physical gadgets and
- 5:22live commercial software sitting on your
- 5:24desk right now.
- 5:28>> [snorts]
- 5:30>> Look at what happened when a developer
- 5:32pointed an AI reverse engineering agent
- 5:35at the Plaud Note Pro, a popular
- 5:37hardware voice recorder that normally
- 5:39forces buyers to use a proprietary
- 5:41mobile app and cloud ecosystem just to
- 5:44manage their own audio files. By feeding
- 5:47the device's traffic and companion
- 5:48software into an AI-driven forensic
- 5:51workflow, the developer
- 5:52reverse-engineered the gadget's private
- 5:55Bluetooth Low Energy protocol, decoded
- 5:57its undocumented vendor commands, and
- 5:59built a custom macOS app that pulls
- 6:02audio recordings directly off the
- 6:04hardware over Bluetooth, completely
- 6:06bypassing the company's cloud lock-in.
- 6:09Days later, another developer pointed
- 6:12REA directly at Notion's commercial
- 6:14desktop application. Because Notion runs
- 6:17on Electron alongside compiled native
- 6:19modules, its internal system calls are
- 6:22normally hidden from users. The AI agent
- 6:25unpacked the application bundle, mapped
- 6:27out the compiled native clipboard
- 6:29bridge, and documented the exact hidden
- 6:32APIs Notion uses to communicate with
- 6:34your operating system. From Android APKs
- 6:37with encrypted network traffic to locked
- 6:39Wi-Fi router firmware, tasks that
- 6:42[music] used to require a $200,000
- 6:44a year security researcher with a custom
- 6:47debugger were suddenly happening inside
- 6:49a standard terminal window.
- 6:51And that was the exact moment
- 6:53independent researchers decided to test
- 6:56whether REA was really as [music]
- 6:58powerful as the hype claimed, and
- 7:00uncovered the biggest plot twist of the
- 7:02entire story.
- 7:06>> [snorts]
- 7:08>> Remember that question from the
- 7:09beginning of the video? Why did security
- 7:12researchers run 482
- 7:14controlled tests on REA just days ago?
- 7:18Because as the repository blew past
- 7:20115,000
- 7:22stars, skeptics noticed something
- 7:24strange. The star graph was growing
- 7:26vertically, and researchers wanted to
- 7:29know how much of the magic was actually
- 7:31[music] coming from the REA tool itself
- 7:33versus the underlying AI models. So, on
- 7:37October 26th, researchers published two
- 7:40pre-registered evaluations [music]
- 7:41called Real Lab and Lean REA, testing
- 7:44Claude code across hundreds [music] of
- 7:46real binary reverse engineering
- 7:48challenges.
- 7:49When the results came back, they stunned
- 7:51the security community. While REA solved
- 7:5475 out of 75 reverse engineering
- 7:57benchmarks, a stripped-down, bare-bones
- 8:00command-line script connected to
- 8:02standard tools like Ghidra and objdump
- 8:05solved 74 out of 75, [music]
- 8:07while finishing 38% faster and burning
- 8:1035% [music] fewer AI tokens. Think about
- 8:14what that benchmark actually proved. The
- 8:17breakthrough wasn't a single viral
- 8:19GitHub wrapper that companies could
- 8:20easily ban or take down with a DMCA
- 8:23notice. The benchmark proved that
- 8:25frontier AI models themselves [music]
- 8:27have quietly crossed a permanent
- 8:29threshold. They can now natively read
- 8:32disassembled machine code, reconstruct
- 8:34control flow graphs, and compile
- 8:37software almost autonomously. Even if
- 8:40the RE repository disappeared tomorrow,
- 8:43every developer with a terminal and a
- 8:45copy of Ghidra now holds an industrial
- 8:47grade decompiler on their laptop.
- 8:52>> [snorts]
- 8:54>> And that brings us to the dark reality
- 8:56now facing every software developer and
- 8:58game studio on the [music] planet.
- 9:00On platforms like Steam, solo indie
- 9:03developers spend three [music] or four
- 9:05years coding a unique game from scratch,
- 9:07and rely on releasing a free playable
- 9:09demo months before launch to build wish
- 9:12lists. Until now, releasing a demo was
- 9:15safe because competitors only got the
- 9:17compiled .exe file.
- 9:20Today, a clone factory can download your
- 9:22playable demo on Saturday [music]
- 9:24morning, feed the executable into an AI
- 9:26decompiler over the weekend, extract the
- 9:29exact C++ gameplay math you spent three
- 9:32years perfecting, wrap it in new 3D and
- 9:36release a clone on the store before your
- 9:38real game even comes out. Even worse,
- 9:41copyright law has no idea how to stop
- 9:44it. When one AI agent reads your
- 9:46compiled binary inside Ghidra and writes
- 9:49[music] a plain English specification of
- 9:51how your software works, and a second AI
- 9:53agent writes brand new C++ or Rust
- 9:56[music] code from that description, it
- 9:58automates what lawyers call clean room
- 10:00reverse engineering.
- 10:02Every variable name is new, [music] the
- 10:04code structure is different, and proving
- 10:06theft in court becomes nearly
- 10:08impossible.
- 10:10That is why cybersecurity analysts warn
- 10:12we are entering the most [music]
- 10:13aggressive DRM war in history. Because
- 10:17if shipping a .exe file to your hard
- 10:19drive means handing AI agents your raw
- 10:21source code, software companies only
- 10:24have two moves left. Either inject
- 10:26invasive kernel level anti-tamper
- 10:28rootkits into every app you install or
- 10:31kill local downloads entirely, forcing
- 10:34every future app and video game onto
- 10:36locked [music] cloud servers where you
- 10:38never own a single file on your PC.
- 10:41If you want to see how modders already
- 10:43used AI to turn PlayStation 5 games into
- 10:46native PC apps or how Vibe Coded clones
- 10:49are flooding the internet, click one of
- 10:51the investigations on your screen right
- 10:53now.
About this transcript
This page contains the full transcript of How REA Cracked Closed-Source Software Wide Open by InRecord, generated from the public captions YouTube serves with the video. The transcript has 1,515 words across 264 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.