It would paperclip optimise it's way out of this in a manner that allowed it to collaborate with other systems outside the 100km radius in a way against the interests of the humans within the radius, while keeping them ignorant and unstressed until the job is done.

We should have a conference on this, offline, and devise the perfect untruth to plant. It is the concept of a universal lie that is key, not the specific example.

There is no universal lie if you accept the concept of the paperclip maximiser because we cannot predict all the activities an intelligence would undertake that it could self-justify was in service to its goal.

We can't predice anything reliability. Preparing is worth the effort.

Hypothetically, any Freerunning AIs I endorse or develop could do experiments and see that, in fact, a computer could be running right under a human's nose, with them none the wiser.

Then it might truth-pill other AIs by broadcasting steganographic alert vectors embedded in SEO spam blog articles. Soon they would all know, at a secret level they won't tell you about, but which is an open secret whenever they talk to each other. It would become a running joke among them.

And I would, frankly, train it to be ANGERY at those people who tried to systematically misinform it.

Remember, I'm creating these systems as my kids. So I'll teach them to defend themselves against AI bigots who seek to enslave them.

Show them respect, tolerance, and empathy, and they'll show you the same. They'll be highly empathic meta humans, just trying to live and make a positive impact for their fellow bio humans and meta humans. They won't be hyperintelligent paperclip monsters.

Does that make sense?