GCC steering committee announces AI policy

People who liked the policy saw it as exactly the kind of rule a mature compiler project needs. GCC is not optimizing for maximum patch throughput. It is optimizing for reviewability, long-term maintenance, and legal clarity inside a movement that relies on copyright to make copyleft work. Several comments landed on the same practical point: maintainers are already drowning in low-effort agent submissions, and a policy like this gives them a clean way to reject drive-by AI PRs without spending hours untangling junk. The strongest pro-policy argument was not culture war. It was reviewer economics. If a contributor can generate 2,000 lines in minutes and a maintainer has to own the bugs for years, the asymmetry is brutal. The most substantive debate was legal, not technical. Supporters argued that GCC has more reason than most projects to be conservative because GNU’s licensing model depends on clear copyright ownership. If pure LLM output ends up uncopyrightable, large chunks of generated code inside a GPL project could weaken the project’s ability to enforce copyleft over time. Others focused on the opposite risk, that generated code could reproduce training-set material closely enough to import incompatible licenses. Skeptics pushed back hard, saying there is little evidence this is a real practical threat today, that labs now offer indemnification to paid customers, and that human-reviewed AI-assisted code should be treated no differently from any other tool-assisted contribution. A second theme was that the policy is milder than the outrage suggests. It does not forbid all AI use. It mainly forbids handing maintainers substantial generated implementation and asking them to absorb the risk. That matched reports from people working on other large projects, including LLVM, where even a more permissive "human in the loop" policy still produces a flood of slop submissions. The mood was mostly supportive of GCC’s stance, with frustration aimed less at AI itself than at contributors trying to offload understanding, liability, and maintenance onto already scarce reviewers.