> audit all the data and code that is associated

For a while now there has been very little incentive for providing these alongside the paper, and I don't see why exactly 'AI' would change this. I could even see how making it vague to be harder to test with LLMs could be profitable for citation hackers.