I'm referring to any policy actually; it cuts both ways. I'd actually suggest a per-model policy - e.g. personally I'd have much less doubts accepting Fable/Mythos patches than Sonnet ones. this is obviously unworkable, impossible to validate and excludes mixed-model workflows, but in spirit the 'smart' models/model mixes should be treated differently than the dumb ones. another case is one-liner (approximately) fixes - I wouldn't even bother to self-report these as AI generated.
