The past four weeks have been something of a wake-up call for the video game industry. You’ve read the headlines, so you know the deal by now: AI advancements spearheaded by Claude Opus 5.5 have turned the world of coding on its axis. Decompilations that used to take months, or even years, are now being pumped out on a daily basis, and publishers can’t keep up with the deluge of vibe-coded browser ports and “native” PC recompilations.
I’ve been covering and researching this over the past month, and there’s one recurring topic that keeps popping up in every comment section I read: karma. For many, this is the video game industry’s “you reap what you sow” moment, proof that the AI tools many publishers have brazenly admitted to using are a double-edged sword.
Hasn’t there already been precedent that copyright doesn’t cover AI generated works? So how to these companies plan on legally selling their games, when it’s actually legal to just copy and share them?
Technically speaking, a cleanroom reverse-engineered program isn’t illegal (this is how IBM PC clones were allowed to exist), at least in the US.
On the other hand, since AI cannot hold a copyright, the prompters cannot sue if, say, Sony or Microsoft just take their code and sell it.
Most decompiled games need a copy of the original game for assets (graphics, textures, etc) since those wouldn’t be recreated via AI, and shipping with them would be copyright infringement.
Requirement for “cleanroom” development" are very strict. You’d have to disregard anyone who could have had just the tiniest plausible chance of having seen the tiniest part of the code.
Any LLM usage would invalidate the cleanroom premise.
And more to the point, the cleanroom scenario were some people producing an ‘interface specification’ working backwards from the binary, not decompiling it. If they had decompiled the source code and passed that on, that would not be cleanroom.
This also hinges upon the ‘cloned’ bit being an ABI/API, which by itself is not considered substantial enough for copyright protection.
So the concept of an AI decompilation of a whole application being considered cleanroom and safe is a bizarre concept that seems to spring from wishful thinking rather than anything vaguely realistic.
It should invalidate it, I’m not sure it will.
no im talking about the reverse, of microsoft using ai to make their new games. further in the article.
The critical thing in the IBM BIOS case is that the ABI surface was deemed too trivial to be worthy of copyright protection, and the implementation was credibly not cloned.
When you get to the point of the interface presented to human, I wager there would be consensus that is substantiave enough content there. Also an actual decompile would not be “clean room”.
Purely generated AI content can’t be copyrighted.
However, mix some human effort in, the combined work can be copyrighted. In theory if you can prove a specific thing is purely AI output, it is fair game. Good luck doing so with proof.
The cases cited in this argument all had the would be rights holder stating flat out it was untouched by human hands. Once they did anything to it, the result would be “transformative” and the AI ness won’t count.
Also in this case, the games being decompiled were pre LLM. An LLM clone doesn’t spare the clone of legal issues. A company tried this argument that the knock off legal papers were AI and this fair game and they lost. When Getty sued, the counter argument was that output could be infringing, but it would be the responsibility of the operator.
Conveniently for the rich people, they have rights to make knockoff machines by ingesting everything without permission, but copyright of course still counts for their purposes.
i was talking about the new ai games being generated by microsoft
Then refer to my first point, a pure AI generated output is likely to be uncopyrightable (the two commonly cited cases are the one where someone tried to assert copyright over photos taken by primates, and another where someone tried to assert copyright over music that was generated by a computer without human curation).
However, once they assert some transformative human involvement is mixed in, the combined work is copyrightable. So they do some token human manipulations of slop, keep the knowledge of what is slop and what is treated/human private and confidential, poof, copyright of ‘lightly treated slop’. Similar to how the little mermad generally is public domain, however Disney’s Ariel is copyrightable, despite originating from material exempt from copyright.
Trademark