Why is it so hard to request these AI companies to provide the sources used for their dataset?

Then if most are illegally acquired, seized the models. The AI company could retrain a model from scratch but with legally acquired data.