Should charge AI for training on top of it or get them to donate. A small amount can fund them easily.
Papers submitted to arXiv under its most permissive license should always be free, as in beer, speech, freedom. For researchers that contribute to it, that is the intention for a reason. It is to serve public and corporate good without restriction.
This isn't me siding with AI companies by the way; it's a slippery slope argument.
as if they would pay.... they would pirate the contents as they already did
That would be a trap. It's healthier for a non-profit to have many small funders than a few large ones.