Open source GZDoom community splinters after creator inserts AI-generated code (arstechnica.com)

The long-running GZDoom open-source Doom port has split after a sizable group of contributors announced a fork called UZDoom, citing disagreements with project lead Cristoph Oelckers (Graf Zahl). The immediate trigger was Zahl’s admission that he had inserted untested, ChatGPT‑generated code into GZDoom—calling one snippet “This is what ChatGPT told me for detecting dark mode on Linux.” Developers led by Nash Muhandes said the fork is intended to preserve the GZDoom legacy while adopting a more transparent, collaborative development model. Beyond fan‑community drama, the episode highlights practical and governance challenges for AI-assisted coding in mature, safety‑critical or long-lived open‑source projects. Untested, AI‑authored fragments can introduce platform‑specific regressions, subtle logic errors, or integration brittleness (e.g., how Linux dark‑mode detection should interact with existing rendering and config subsystems). The incident underscores the need for provenance, mandatory code review, CI/tests, and clear policies about AI-generated contributions—issues that will matter to many ML/AI tool users and maintainers as these assistants become commonplace in codebases.