Ask HN: If LLM reverse engineering is deemed illegal, is that precedent for OSS?
With the efforts to clamp down on LLM-assisted reverse engineering - I am wondering;
If LLM assisted reverse engineering is deemed illegal by the courts, is that precedent for a class action that asserts OSS licenses are violated by labs that trained LLMs on that code?
There are two modalities for this;
1) LLM being post-trained against a decompiled binary - then used to output working source code.
2) LLM being used to analyze decompiled binary and used to output working source code.
How can it be "clean room" when an LLM is trained on or reads OSS but not clean room when the same is done for decompiled software?
(
10 comments
| on HN
)
Can an LLM be "clean room"? Are weights trained on decompiled code considered "tainted"?
Trade secrets have no legal protection
Sandy Monroe is a great current example, tears down cars and builds reports. The auto industry has distillation built in
However, the intent to reverse engineer without source--only public API and analysis of data streams--for commercial use almost certainly violates OSS, since it's to evade strictures.
Private use like retrocomputing, "right to repair," should support the individual.
Basically, if LLM's output is relicensed for different intent, courts should see through that.
Some of the reverse engineering being done by LLMs is using older OSS. I remember hearing about some 2012 era reverse engineering app that looks at the assembly language and reconstructs software. The LLM isn't reverse engineering itself in those scenarios, its calling local software, interpreting results, probably glueing software together with what the human is prompting it to do.