Privacy policies actually use this as a rhetorical trick to confuse users. E.g., they'll say stuff like "we may collect any device information you give permissi

Privacy policies actually use this as a rhetorical trick to confuse users. E.g., they'll say stuff like "we may collect any device information you give permissions to access to do [reasonable thing A], [reasonable thing B], or for any other business purposes. Examples of device information include: [insert list of benign things]" Or, to put emphasis on the important things: "we may (read: WILL) collect ANY device information you give permissions to access to do [reasonable thing A], [reasonable thing B], or for ANY other business purposes. Examples of device information include: [insert list of benign things you should totally ignore]" But here's the problem with the "AI tell us all" approach. If you use ML to mine examples from existing policy texts, you'll end up showing the user all of the benign stuff that the companies put in to reassure users, because that's what the corpus contains. Therefore, your AI will emphasize exactly the wrong types of examples. Also, notice that this isn't even a legalese thing. There are no legal terms of art or long-winding difficult grammatical constructions. It's just plain old good rhetoric. And it's hard/impossible to legislate away persuasion.