Training Artificial Intelligence (AI) requires vast amounts of training data. Without such large corpora these AI would be much less useful. However those hosting content might wish to communicate their preference that this not take place, or that only some AI uses but not others be allowed. This Internet Draft attempts to enable hosters of content served using HTTP (and possible FTP) to express their preferences.

To enable the expression of preferences a directive could be placed in a robots.txt file or in a HTTP header. This directive would specify a category of Artificial Intelligence (see figure) and a preference of either allow or deny. Any unstated preference would be no preference. In the case of two rules being contradictory the more specific category would take precedence. New categories could be added but only as subsets of existing categories as long as the new category wasn't a superset of an existing category.