I've heard plenty of accessibility arguments made on moral grounds. Rarely anything happens because moral grounds don't get a budget.
I'd argue for a different approach.
We can agree that fixing accessibility means audits, contractors, development sprints and retrofitting things you should have built right the first time. That all sounds expensive because it is. It's also slow enough to justify kicking it down the road indefinitely, which is what usually happens.
What if it wasn't as expensive or as slow?
Could AI change that equation?
I imagine error messages rewritten automatically, image descriptions generated for you as a starting point, form labels being flagged before you ship and copy simplified without a specialist. Maybe these aren't hypotheticals. And if they aren't, then the cost drops significantly and the speed increases by quite a bit.
But these things only happen if a few things are true.
Someone has to own this. There's no way AI fixes accessibility by sheer presence in your workflow. Someone still needs to decide accessibility is part of the definition of done. You still need to integrate it into your workflow and not slap it on at the end.
If this sounds like the same arguments with or without AI, it's because they are. AI is a tool like any other. It can reduce cost and can increase speed. But it surely doesn't make the decision for you.
That one still sits entirely with you.