Research paper warns that AI-generated 'slop' is polluting the open-source software commons, making it harder for developers to find reliable code

A new paper (arxiv.org/abs/2604.16754) documents how low-quality AI-generated code is flooding public code repositories and open-source projects, mixing good and bad implementations together in ways that confuse both humans and training algorithms.

Unlike a clearly labeled spam folder, AI-generated code looks legitimate on the surface—it compiles and runs—but often contains subtle bugs, inefficient logic, or security gaps that only fail under specific conditions. This means developers spend hours debugging code that seemed fine at first glance.

For software teams, this means open-source libraries and code samples you copy-paste become less trustworthy; you can no longer assume a GitHub example works correctly. For companies training their own AI models on public code, poisoned datasets mean your AI assistant will learn bad patterns and pass them on to your engineers.