Surely they should have to comply with all the licenses covering the training set, e.g. GPL3. So not just open-source, but copy-left, anti-tivoisation and all of that.
All AI-generated source should be ineligible for copyright protection at all, just like is the case for AI-generated art.
But guess how that’s going to pan out when it gets to court.
Given that 90+% of commercial software is written by or with huge support of the clanckers these days, I highly doubt this. We are far beyond the point at which this would have been a possible decision.
At least at the moment in the US, the copyright office makes no distinction between visual art and LLM slop, in that neither is eligible for protection. The courts have so far upheld this, though there haven’t been any “blockbuster”-level cases yet.
So for the moment, AI-generated code is ineligible for protection.
Got to thank that monkey selfie for that ruling from the copyright office.




