Generally creating a lot of friction around contributing to a project defeats the whole purpose of open sourcing it.
A better policy might simply be to automatically (using AI for this is kind of an obvious thing to do) filter and classify incoming PRs and issues for complying with quality thresholds.
A huge PR that comes in without a clear discussion history in the issue tracker is kind of a rude thing to do. That should be an automatic close. Have some good contributor guidelines and then enforce them. With or without AI. Block repeat offenders that can't be bothered to stick to those. Most of the annoyance comes from having to do all this manually and getting distracted by all this noise.
A small, focused fix that comes in with a well articulated explanation of what was done and why is a different matter. It shouldn't matter if the contributor used AI or not.
The main issue is that the signal is drowned out with the noise with a lot of problematic contributions from new/unverified users. But those should be kind of easy to detect as well.
There are multiple ways to deal with this. But a blanket ban on AI is a bit throwing out the baby with the bath water.
I actually have the opposite problem on my OSS repositories. It seems people are to busy doing their own projects to actually open pull requests on my projects. There's a noticeable decline in the number of pull requests since last year.