A complete mischaracterization, as usual for HN lately when discussing AI or LessWrong. Obviously, even average levels of persuasion are enough to convince some people. And nobody is air-gapping AI.
https://rationalwiki.org/wiki/AI-box_experiment
https://www.yudkowsky.net/singularity/aibox
https://www.lesswrong.com/posts/Bnik7YrySRPoCTLFb
It's far from the craziest belief that's come out of that group.
I mean... I would hope that AI used for military needs is not deployed in the public internet.