This is a pretty good middle ground, I think. You can't prevent LLM usage and there's significant downsides to doing so universally, so restricting contributions to things that a human needs to demonstrably understand circumvents a lot of problems.