I was wondering what this "clean-room" jargon actually meant, after looking at the Adobe ripoffs being vibe coded by ArtCraft.
As with most forms of theft and laundering, it's not really that clever.
>> This document records behavior established from static analysis of the retail TotalA.exe. It is a clean-room behavioral contract, not a source translation. It deliberately omits executable addresses, data-image offsets, and raw decompiler names. The phrases established, supported inference, and unknown distinguish what the analysis proves from what remains a useful but unverified interpretation.
The evidence is the executable's own control and data flow, its PE import table, and its embedded string vocabulary. Nothing here is taken from another engine, a capture, or a reimplementation.
Step 1: Ask an LLM to decompile a binary, wash it into a "clean" description of the binary in its own format (a bunch of .MD files, one might suppose) and then build a new binary from that description.
Step 2: Nothing.
Is this really going to hold up in court?
IIRC, that's how Compaq cloned IBM's PC BIOS, with humans doing the analysis instead of an LLM, without leaving IBM any legal grounds to sue Compaq for the cloning. Being a clean room re-implementation is also why Linux was able to clone UNIX without getting sued out of existence by AT&T or SCO so we'd all jolly well better hope that it still provides legal protection.
If they had sent the machine code through a modem, out a speaker, into a microphone and back into a modem, that would've been illegal. So what you're talking about, really, is just a legal exemption that's been carved out for humans who do work from scratch themselves. There's nothing about cloning a BIOS with a team of people rebuilding it that is necessarily not IP theft, or legally sacred either. It's just that there's an even stronger case to be made that having an LLM do it is fundamentally not clean, because the LLM is itself just a massive decompiler / compiler. A bunch of humans performing the same action are somewhat exempt by custom.
Are you using the word "theft" colloquially or literally?
I'm just curious, people consider this theft?
Yeah? If I print out a pdf and then scan it back in, it doesn't matter what process was followed to end up with an identical end product.
imo the better question is whether it will actually get to court, and the precedent on that is mostly no. so it doesn't really matter i think
Copyright protects expression. Good luck with a technology that can morph to a hundred equivalent expressions.
I mean, they’ll probably train a law LLM to find a way to attack it.
If you want to make a lot of money, this is the way.