The protocol as it stands has a very obvious attack vector; it would be extremely easy for an AI slop farm to start providing their own human.json files, and for someone outside of the slop farm to unwittingly (or coercively) put a link into it from their own file. A web of trust also requires the ability to add a "web of distrust," so that people can flag a site as being AI-written and potentially flag someone else's human.json file as being inaccurate in that regard.
This would also pair with what I mentioned in #53 with providing evidence.
The protocol as it stands has a very obvious attack vector; it would be extremely easy for an AI slop farm to start providing their own human.json files, and for someone outside of the slop farm to unwittingly (or coercively) put a link into it from their own file. A web of trust also requires the ability to add a "web of distrust," so that people can flag a site as being AI-written and potentially flag someone else's human.json file as being inaccurate in that regard.
This would also pair with what I mentioned in #53 with providing evidence.