The problem with conspiratory talk is that while one person may fully not intend on action, it could inspire and/or manipulate others into committing acts. The blame is shared on all for conspiring and creating that environment where acts can emerge.
What if an unhinged language model generates all this noise talking to other language models, with no humans involved at all? The only human involvement would be an instruction to start spouting some believable bullshit on controversial topics, plus granting access to some private messaging tools and providing a contact list of other language models to talk to.
The problem with conspiratory talk is that while one person may fully not intend on action, it could inspire and/or manipulate others into committing acts. The blame is shared on all for conspiring and creating that environment where acts can emerge.