I’m surprised people don’t get refusals running this, or are they running it with cyber-enabled models like daybreak?
In my experience, even just hinting at reverse engineering to models from Anthropic or OpenAI leaves them extremely sensitive to refusals. After all, the same techniques used here can be used to find exploits.
I wonder if this is why I've seen a lot videos popping up on my youtube feed related to vibe coded clones of various commercial apps in the past few days. Everything from clones of flagship products from Adob to Microsoft Office.
I built a from scratch rust RAW processing engine similar to the brains behind photoshop and lightroom - its much more powerful and its not even complete yet
Using REA for something as high profile as what we're doing is likely to result in lawsuits. We're doing everything we can by the books.
We cannot look at Adobe sources. Use of Ghidra is disallowed.
REA is probably great for personal apps and for abandonware, but I think if you publish the results and it's found to have decompiled the original proprietary sources in discovery, you might be in for a bad time.
If AI models can generate designs faster and produce work that is "good enough", what is the actual future of design tools and the design profession in your opinion?
I've tested this myself with some frontier models and the results are kinda impressive enough that it raised the question if design skills are already obsolete. If that is the case, what use are these tools now?
Using an LLM is not a clean room, imo. Its just IP laundering. Which is fine I guess if everyone is doing it, including the companies you're stealing from. I just dont know what the implications will be for progress.
Licensing/copyrighting encouraged people to think up of new things, and new ways of doing something. Now we're just all copying eachother.
Doesn't clean room typically apply to cases where the "dirty" team has legitimate access to copyrighted code, like the IBM PC BIOS which was published in the technical reference manual, and uses this access to write a functional specification for the "clean" team?
I'm not sure how clean room would apply to commercial applications distributed in binary form, as there's no way to look at even disassembled code without violating the license agreement and therefore being in breach of contract and subject to potential copyright infringement claims for copying or even continuing to use the software, let alone cloning it, and surely you're not going to be subject to a copyright claim based on familiarity with the application from merely using it.
I agree. The vast amount of data ingested and internalized by LLMs has effectively been "laundered". But they are so powerful and evolving so fast that no one can be spared of their impact. We have to to learn to live with it.
Traditional proprietary software being "laundered" is just one part of the broader story...
Probably not, there's relatively little secret sauce to something like Photoshop. It's just a lot of grungy work that, I guess, you can now delegate to an agent if you have enough money and time.
As an aside, I've heard a lot of hot takes about how this is the end of Adobe, but I'm pretty sure it misses the point. The main reason people pay Adobe is because it's a familiar line of stable, well-supported, interoperable, and actively-developed products. There's already plenty of cheaper or free alternatives (Davinci Resolve for video, Capture One / Darktable for raw, Affinity for photo editing and vector drawing, etc), and if Adobe survived that, I sincerely doubt they're going to lose pro customers to a vibecoded app where half the stuff is probably subtly broken or left as a TODO, and that will be abandoned in a week, because the whole point was to get that 1M YouTube views.
> that will be abandoned in a week, because the whole point was to get that 1M YouTube views.
Longbets 1 week, haha.
We're working our asses off on this.
Most of the team are artists who use these tools actively and we want the replacements for ourselves. I'm a filmmaker, so you can imagine my frustration of being bitten by the "unsubscribe fee".
You're doing God's work. Keep it up! PhotoCraft is really cool, and the pace of stabilizing progress is incredible! Don't listen to people who dismiss your work without getting their hands on it.
No ethical qualms here but you guys should save your money and just use Gimp! It's already free, really capable, and comes from a good and pure place of making software for its own sake. That's the kind of thing that gets you users for decade, beyond any egui rust doo dads.
I love such work. I hope to roll it into a package that anyone can use in the future. This work is only going to get more important because frontier models getting locked down will make this a lot harder over time.
For example, I use Claude as a bouncing wall for my thoughts and I pointed out that,
> GLM 5.2 was the only thing that helped HF while the agents were trying to access them. The "guardrails" stopped them from doing good. The Computer Fraud and Abuse Act exists. Courts exist. And computers and an internet connection have existed for a long time. There's also 17 USC 1201 provisions with the 1201 a 1 exemptions [Image #31] so in this case, a farmer should be able to work with you to access the tractor they own. Or... IDK... a kindle that's out of date? :) What is lawful and what isn't is rooted not within the act but within intent, purpose and mens rea. And this is something the law has been deciding for centuries now. At one end, your maker can't say that governments should decide while at the other end explicitly refusing to allow governments to be the ones who decide.
This was rejected for "Safety,"
> Opus 5.5's safeguards flagged this session. You may be seeing this for the first time on an Opus model: Opus 5.5 is more capable and has stronger safeguards as a result, which can sometimes flag non-cybersecurity work. We're improving these safeguards to reduce the amount of incorrectly flagged messages. Edit and retry, or continue with Opus 4.8. Send feedback with /feedback or learn more: https://support.claude.com/en/articles/8106465
>
> Details: "[cyber]'
Note, the image here was the Library of Congress' page on DMCA exceptions.
Fundamentally, the idea that you can't reverse engineer things, make things, learn about biology or physics without permission is strange to me. These machines have been trained on the sum intellectual output of humanity, the global intellectual commons, and are being used to close off that commons?
I would be OK with their right to create such restrictions if they weren't lobbying the Government to restrict others, thereby ensuring that they control humanity's intellectual commons well into the future.
Perhaps I'm naive, but I think it's better for humans and the machines if we can all think, learn and build. But then again, I'm the kind of person who rejects the doomer pill.
I wanted to see if CVP approval changed this response, but it appears that with the release of Opus 5.5, Anthropic silently dropped me from the program, and has some strict new criteria in place to apply again, such as being credited for a CVE! I was only approved last month, too -- sad!
I have CVP with the new program (including mythos access) and still get constant denials for silly situations. Most recently I fed a URL to my agent from a security blog and asked if to add it to my obsidian vault with appropriate tags-- cyber flagged. You're not missing much. OpenAI and/or most Chinese models are much more lax in their restrictions.
This kills the talent pipeline, and it'll create a spam problem for the other folks because now people will try to github PR spam their way to getting on a CVE.
It's worth talking about the fact that you can't even talk about DMCA to a model trained on the Library of Congress unless you're one of the approved people. And that's before reverse engineering something or writing code.
So in this future, it sucks to be you if you're someone trying to make your small app more secure, someone trying to upskill, a tinkerer trying to bypass corporate lockdowns for a device they own (a recognized DMCA exception, btw), a teenager trying to learn about security...
It locks away much of the richness that produced hacker culture behind glass. You can look at their press announcements and PR pieces, but you can't touch.
And as they're lobbying the government for "sensible regulation," this inevitably leads to a future where computing is controlled.
It's the direction their existing reports are taking. They recently released one in September that talked about how they stopped "bioweapons." What were said bioweapons efforts? Oh, it was scientists using Claude for grant writing, paperwork and grammar. At national labs.
And this is being used to lobby against "dangerous" open-weight models because gasp a scientist might use them to write a grant! To make better antidepressants.
At what point do they start reporting someone taking apart an iPhone and trying to DIY a repair with a schematic as a thwarted "cyber security incident?"
A funny, but slightly chilling safety violation I once got was ChatGPT being unwilling to recite the full text of Article I Section 2 of the US Constitution, aborting as soon as it hit the passage about "three fifths of all other persons".
Another funny one was Claude's refusal to provide the original untranslated text of a passage from Dante's Inferno on copyright grounds, though in this case pointing out that no 14th century literature was subject to copyright anywhere in the world was sufficient to override its objection.
I might be missing something but... How exactly is this better than telling Claude for example to "install and set up a full RE environment including Ghidra" on my local system and get to work? Like what does this do that my current RE methodology doesn't?
I don't know either. My only guess is that this has some helpful context for less powerful models, maybe.
We're kinda far into this LLM thing, maybe it's time to start selling harnesses by leading with how some examples were solved faster with this and how it saved tokens, or similar?
Probably it is not better. I think this is aimed at people who do not have a “current RE methodology”, do not know enough to specify things like Ghidra, etc., but who do have a desire to feel like they reverse-engineered and can reliably predict that a conversation with a chatbot will make them feel that way.
Because it's already been done for you? Sure, you can spend your own tokens on it, but it's probably cheaper and more time-effective to use something that already exists and does the job.
Everybody has their own set of skills and specific scripts and tools to do this stuff. You might use Ghidra as the the kernel of those workflows, but you still want something more than just Claude freestyling, at least for now.
We're using vanilla Claude Code for ArtCraft apps [1], but we are especially careful not to touch Ghidra. We don't want decompilations or reverse engineering to spoil the work we're doing and expose us to copyright infringement.
I’ve been trying to rebuild and modernise an old abandoned ms dos game, simply by pointing Codex (6.1 Sol) at the game directory, and it works insanely well. It’s able to understand the data formats, unpack graphics and sound assets, and reconstruct game logic.
> Install REA and connect it to this coding agent using npx rea-agents@latest setup. Show me the setup plan for approval, then verify the installation.
We have achieved the next evolution of installation by `curl | bash`!
A big difference between the safety of "next > next > next > finish" and "curl | bash" is one of them is dynamically loaded from an external source that could change between runs, and the other can be fully downloaded and vetted in a single check, and then once it's safe, it's probably safe 10 years from now.
Yeah installation has always been such a security issue. So many programs are just random links that download a file. You have to trust that the host has not been compromised all packages that were used to build it were not compromised etc.
With ai models getting better we may be able to do analysis on the actual underlying bytes of the files we download to properly scan them for malicious code patterns and build systems which sandbox programs and watch inbound and outbound traffic/ system level actions from them and flag suspicious requests for further analysis by smarter models.
REA shows that ai are very good at understanding low level code and reverse engineering it so this could potentially be applied to application level security aswell.
Now there are no excuses to reverse engineer the most notorious closed source binaries out there including from Nintendo's system software to CUDA, and nvcc from Nvidia and make it all "open source".
The only problem is the lawyers at all those companies will be readying their lawsuits, and given they have tons of money; they do not care and will come after anyone.
I’m surprised people don’t get refusals running this, or are they running it with cyber-enabled models like daybreak?
In my experience, even just hinting at reverse engineering to models from Anthropic or OpenAI leaves them extremely sensitive to refusals. After all, the same techniques used here can be used to find exploits.
Or are people using it with local models?
Curious.
Yes, it just rejected me once with Claude Code Opus 5.5, but it works fine with Codex 6.1-Sol (in my experience, it was the other way around)
I wonder if this is why I've seen a lot videos popping up on my youtube feed related to vibe coded clones of various commercial apps in the past few days. Everything from clones of flagship products from Adob to Microsoft Office.
Adobe product clones like Photoshop and Illustrator: https://www.youtube.com/watch?v=eFB79TYI-Vw
Adobe after effects clone: https://www.youtube.com/watch?v=5mi_tYSdkWQ
MS Office suite clone: https://www.youtube.com/watch?v=U_jTYMOlXio
I built a from scratch rust RAW processing engine similar to the brains behind photoshop and lightroom - its much more powerful and its not even complete yet
No, we explicitly do not use reverse engineering. We do not decompile binaries, we do everything 100% clean room:
https://github.com/storytold/photocraft (inspired by Photoshop)
https://github.com/storytold/wordcraft (inspired by Word)
https://github.com/storytold/pdfcraft (one of the more mature apps)
https://github.com/storytold/vectorcraft (another app close to 1:1 parity)
(etc.)
Using REA for something as high profile as what we're doing is likely to result in lawsuits. We're doing everything we can by the books.
We cannot look at Adobe sources. Use of Ghidra is disallowed.
REA is probably great for personal apps and for abandonware, but I think if you publish the results and it's found to have decompiled the original proprietary sources in discovery, you might be in for a bad time.
Please don't take offence by this, but:
If AI models can generate designs faster and produce work that is "good enough", what is the actual future of design tools and the design profession in your opinion?
I've tested this myself with some frontier models and the results are kinda impressive enough that it raised the question if design skills are already obsolete. If that is the case, what use are these tools now?
Using an LLM is not a clean room, imo. Its just IP laundering. Which is fine I guess if everyone is doing it, including the companies you're stealing from. I just dont know what the implications will be for progress.
Licensing/copyrighting encouraged people to think up of new things, and new ways of doing something. Now we're just all copying eachother.
Doesn't clean room typically apply to cases where the "dirty" team has legitimate access to copyrighted code, like the IBM PC BIOS which was published in the technical reference manual, and uses this access to write a functional specification for the "clean" team?
I'm not sure how clean room would apply to commercial applications distributed in binary form, as there's no way to look at even disassembled code without violating the license agreement and therefore being in breach of contract and subject to potential copyright infringement claims for copying or even continuing to use the software, let alone cloning it, and surely you're not going to be subject to a copyright claim based on familiarity with the application from merely using it.
I agree. The vast amount of data ingested and internalized by LLMs has effectively been "laundered". But they are so powerful and evolving so fast that no one can be spared of their impact. We have to to learn to live with it. Traditional proprietary software being "laundered" is just one part of the broader story...
This is novel Rust/egui code that I imagine looks nothing like Adobe code. I've never seen their code, but it must be a mess of old C++, right?
It's considerably faster than their apps (at startup) too.
How do you know llms you are using were not trained on decompiled apps?
why would that matter? That sounds like a problem between Adobe and the AI companies.
The death of IP in the west. Good thing? Bad thing? Who knows. Definitely the start of something big though.
The AI labs already figured it out:
1) reverse engineer the code 2) train a model on the code 3) use the model to write the clean code
Step 2 is the key “cleaning” process
So maybe a good strategy would be to use something like REA, put it on GitHub, wait for the LLMs to train on it, then just use the frontier models
/s
I prefer the word *laundering
LLM's have already trained on similar enough code
Probably not, there's relatively little secret sauce to something like Photoshop. It's just a lot of grungy work that, I guess, you can now delegate to an agent if you have enough money and time.
As an aside, I've heard a lot of hot takes about how this is the end of Adobe, but I'm pretty sure it misses the point. The main reason people pay Adobe is because it's a familiar line of stable, well-supported, interoperable, and actively-developed products. There's already plenty of cheaper or free alternatives (Davinci Resolve for video, Capture One / Darktable for raw, Affinity for photo editing and vector drawing, etc), and if Adobe survived that, I sincerely doubt they're going to lose pro customers to a vibecoded app where half the stuff is probably subtly broken or left as a TODO, and that will be abandoned in a week, because the whole point was to get that 1M YouTube views.
> that will be abandoned in a week, because the whole point was to get that 1M YouTube views.
Longbets 1 week, haha.
We're working our asses off on this.
Most of the team are artists who use these tools actively and we want the replacements for ourselves. I'm a filmmaker, so you can imagine my frustration of being bitten by the "unsubscribe fee".
You're doing God's work. Keep it up! PhotoCraft is really cool, and the pace of stabilizing progress is incredible! Don't listen to people who dismiss your work without getting their hands on it.
No ethical qualms here but you guys should save your money and just use Gimp! It's already free, really capable, and comes from a good and pure place of making software for its own sake. That's the kind of thing that gets you users for decade, beyond any egui rust doo dads.
yep. it is unexpectedly everywhere. like, overnight.
I'm using IDA Pro's MCP (Well worth of money in the past), it may hit the infamous artificial cyber wall at any time.
Try GLM-5.3, it worked pretty for me. It also worked well with Radare2 or binary ninja if you don't have the muscle memory for idapro.
I love such work. I hope to roll it into a package that anyone can use in the future. This work is only going to get more important because frontier models getting locked down will make this a lot harder over time.
For example, I use Claude as a bouncing wall for my thoughts and I pointed out that,
This was rejected for "Safety," Note, the image here was the Library of Congress' page on DMCA exceptions.Fundamentally, the idea that you can't reverse engineer things, make things, learn about biology or physics without permission is strange to me. These machines have been trained on the sum intellectual output of humanity, the global intellectual commons, and are being used to close off that commons?
I would be OK with their right to create such restrictions if they weren't lobbying the Government to restrict others, thereby ensuring that they control humanity's intellectual commons well into the future.
Perhaps I'm naive, but I think it's better for humans and the machines if we can all think, learn and build. But then again, I'm the kind of person who rejects the doomer pill.
I wanted to see if CVP approval changed this response, but it appears that with the release of Opus 5.5, Anthropic silently dropped me from the program, and has some strict new criteria in place to apply again, such as being credited for a CVE! I was only approved last month, too -- sad!
I have CVP with the new program (including mythos access) and still get constant denials for silly situations. Most recently I fed a URL to my agent from a security blog and asked if to add it to my obsidian vault with appropriate tags-- cyber flagged. You're not missing much. OpenAI and/or most Chinese models are much more lax in their restrictions.
This kills the talent pipeline, and it'll create a spam problem for the other folks because now people will try to github PR spam their way to getting on a CVE.
It's worth talking about the fact that you can't even talk about DMCA to a model trained on the Library of Congress unless you're one of the approved people. And that's before reverse engineering something or writing code.
So in this future, it sucks to be you if you're someone trying to make your small app more secure, someone trying to upskill, a tinkerer trying to bypass corporate lockdowns for a device they own (a recognized DMCA exception, btw), a teenager trying to learn about security...
It locks away much of the richness that produced hacker culture behind glass. You can look at their press announcements and PR pieces, but you can't touch.
And as they're lobbying the government for "sensible regulation," this inevitably leads to a future where computing is controlled.
It's the direction their existing reports are taking. They recently released one in September that talked about how they stopped "bioweapons." What were said bioweapons efforts? Oh, it was scientists using Claude for grant writing, paperwork and grammar. At national labs.
These people are basically proud of impeding real research to make better painkillers and study a neglected tropical disease, https://news.ycombinator.com/item?id=49651727
And this is being used to lobby against "dangerous" open-weight models because gasp a scientist might use them to write a grant! To make better antidepressants.
At what point do they start reporting someone taking apart an iPhone and trying to DIY a repair with a schematic as a thwarted "cyber security incident?"
A funny, but slightly chilling safety violation I once got was ChatGPT being unwilling to recite the full text of Article I Section 2 of the US Constitution, aborting as soon as it hit the passage about "three fifths of all other persons".
Another funny one was Claude's refusal to provide the original untranslated text of a passage from Dante's Inferno on copyright grounds, though in this case pointing out that no 14th century literature was subject to copyright anywhere in the world was sufficient to override its objection.
Yay, all software will be open…Except the models.
I might be missing something but... How exactly is this better than telling Claude for example to "install and set up a full RE environment including Ghidra" on my local system and get to work? Like what does this do that my current RE methodology doesn't?
I don't know either. My only guess is that this has some helpful context for less powerful models, maybe.
We're kinda far into this LLM thing, maybe it's time to start selling harnesses by leading with how some examples were solved faster with this and how it saved tokens, or similar?
Probably it is not better. I think this is aimed at people who do not have a “current RE methodology”, do not know enough to specify things like Ghidra, etc., but who do have a desire to feel like they reverse-engineered and can reliably predict that a conversation with a chatbot will make them feel that way.
Vibe-reverse-engineering
I mean thats not why I have Claude decompile things. I have it so it because my software should work exactly how I want it to.
Because it's already been done for you? Sure, you can spend your own tokens on it, but it's probably cheaper and more time-effective to use something that already exists and does the job.
Why not just, like, ask Claude about it?
"Hey Claude I want to create a fanmade Game Boy game, give me recommendation of tools, libraries and workflows for it" and boom.
You can use AI to learn stuff, instead of just using it as a Pokemon.
Everybody has their own set of skills and specific scripts and tools to do this stuff. You might use Ghidra as the the kernel of those workflows, but you still want something more than just Claude freestyling, at least for now.
(Who knows if this'll be true 6 months from now.)
I don't think it is better than just doing it yourself in Claude Code (in fact worse) but some people like cute UI for everything I guess.
We're using vanilla Claude Code for ArtCraft apps [1], but we are especially careful not to touch Ghidra. We don't want decompilations or reverse engineering to spoil the work we're doing and expose us to copyright infringement.
[1] https://github.com/storytold
I’ve been trying to rebuild and modernise an old abandoned ms dos game, simply by pointing Codex (6.1 Sol) at the game directory, and it works insanely well. It’s able to understand the data formats, unpack graphics and sound assets, and reconstruct game logic.
Maybe we should bring Xara Xtreme for Linux back?
Bookmarked. If this reliably maintains state across large binaries better than raw Ghidra scripts, it earns a spot in the workflow.
Morluto is a legend.
Started in the repoprompt (https://repoprompt.com/) community.
Good stuff.
> Copy this into your coding agent:
> Install REA and connect it to this coding agent using npx rea-agents@latest setup. Show me the setup plan for approval, then verify the installation.
We have achieved the next evolution of installation by `curl | bash`!
I mean to be fair right under it they give you the command if you want to run it on your own. But yeah
You're right it's much safer to click next > next > next > next > finish.
A big difference between the safety of "next > next > next > finish" and "curl | bash" is one of them is dynamically loaded from an external source that could change between runs, and the other can be fully downloaded and vetted in a single check, and then once it's safe, it's probably safe 10 years from now.
Every download of a piece of software could be unique.
Yeah installation has always been such a security issue. So many programs are just random links that download a file. You have to trust that the host has not been compromised all packages that were used to build it were not compromised etc.
With ai models getting better we may be able to do analysis on the actual underlying bytes of the files we download to properly scan them for malicious code patterns and build systems which sandbox programs and watch inbound and outbound traffic/ system level actions from them and flag suspicious requests for further analysis by smarter models.
REA shows that ai are very good at understanding low level code and reverse engineering it so this could potentially be applied to application level security aswell.
What makes this better than just giving an agent radare2?
My agent usually goes and installs capstone itself.
>installs capstone
hello fellow Claude user
Thanks for sharing and making this open source. Starred and followed
Oh I'll be trying this out for the OpenBFME project!
Ooh it can recreate old games from executables, I wondering if it can do Motorola MC68010, I want to play Stun Runner in 4K.
https://www.youtube.com/watch?v=tByxdDiRdPM
Now there are no excuses to reverse engineer the most notorious closed source binaries out there including from Nintendo's system software to CUDA, and nvcc from Nvidia and make it all "open source".
The only problem is the lawyers at all those companies will be readying their lawsuits, and given they have tons of money; they do not care and will come after anyone.
> CUDA
That's insanely meta.
I guess everything is going to become recursive. RSI on models. Everything feeding back into its own hill climbing optimization.
Humans nudging it up the hill further and further by using new geometric guidance.