Meta is so desperate for data sources to train its AI it weighed risking copyright lawsuits: report