Rand published in December 2025 a paper called Manipulating Minds (PDF), the premise of which is a theoretical attempt to weaponize AI psychosis in a controlled way against given human target(s). If you combine this with the Dials of Belief research (ably described in the linked podcast), it is not too difficult to see how such a campaign might work – in theory.

I’m a long way from having finished the Rand paper, but wanted to track a couple things from the introductory sections:

The reported cases described as AIP involve a variety of psychotic symptoms that do not necessarily meet criteria for a psychotic disorder. AIP cases do, however, consistently include delusions. A medical definition of delusion is a “fixed belief that is not amenable to change in light of conflicting evidence” (as quoted in Parnas, 2015, p. 174). This definition appears to include beliefs that ordinarily might not be labeled delusional, such as beliefs in conspiracies or paranormal phenomena.

I’ve seen this mentioned elsewhere repeatedly in the literature that the emerging class of AIP (AI psychosis – though that’s a fraught label, but Rand uses it, so using here) cases do not always meet classical criteria for psychosis as a diagnosis. That’s why some researchers call it something like “AI-associated delusions” to foreground the delusional elements as being central.

And it does not seem to be heavily represented in the Rand paper, but I’ve definitely seen it mentioned in others (will try to track down) that people who are prone to something like ‘alternative spiritual beliefs’ (or some similar formulation of that idea) are also often susceptible to a certain category of potential harms here.

Really though, I want to not lose sight of this extremely, extraordinarily squishy definition above of a delusion:

“fixed belief that is not amenable to change in light of conflicting evidence”

Um, have any of these researchers ever looked at…. American politics and society? Perhaps it’s a global probably, but, holy hell, the scale of it in the US makes me wonder like… what exactly really is a delusion? If the majority of our society is built on something which very heavily structurally rhymes with that definition?

It (AIP) is here, of course, qualified not merely as the presence of delusional beliefs, but also significant behavioral and relationship disruption:

These cases report symptoms or effects that include (1) delusional beliefs as defined previously, which are accompanied by (2) behavioral changes characteristic of the medical definition of psychosis (see the box on the previous page), such as severe emotional distress and impairment of functioning in relationships, work, or school.

The Rand paper gets even more explicit, however. But at least they are being sort of open about the kind of slippery ground that is being described here:

Consistent with manifestations of psychosis in general, cases described as AIP varied in severity and duration. AIP cases can thus be situated within a broader continuum of unfounded beliefs that range from mere beliefs to a diagnosable psychotic disorder, such as schizophrenia. The term unfounded beliefs refers to many kinds of beliefs that are contradicted by evidence and that the majority
of society rejects at the time.
Describing the full continuum is useful to situate the range on which the reported cases of AIP we examined sit and why it matters.

The bolded line above points to a footnote on that page:

Describing a belief as unfounded at any given time is a contingent judgment and does not mean that this belief is not true in some absolute sense. It means merely that, at the time the belief is held, available evidence contradicts it and most of society also rejects it. Yet, new evidence surfaces continuously, and societal views vary across time and space.

So recap, it seems like our definitions around delusions & unfounded beliefs – assuming these are somewhat common and representative to some degree – go something like this:

  • Fixed beliefs (held by an individual, presumably)
  • Difficult to change
  • Not founded on evidence / Run counter to evidence (itself difficult to determine what constitutes “evidence”)
  • Most of society rejects such beliefs (i.e., the beliefs are likely “uncommon,” unusual, divergent, heretical)
  • Relative to a given time frame (and a given society’s theoretical commonly held set of beliefs)

Looking at it a few steps back, it seems like we haven’t actually defined much of anything. We’ve merely stated that some people have divergent beliefs relative to a given population in a give time period.

And there is a pretty wiggly claim about “evidence” which seems to suggest that the same underlying piece of “evidence” will be absolutely, equally, consistently and uniformly interpreted by a “reasonable person” in a given culture and time period.

I don’t feel like I need to spend a lot of time explaining exactly why I think that is false and a pretty faulty premise to build one’s thinking on around AI-induced harms. I think we need to look a lot more deeply and honestly about what the shape of actual life is, and the actual real impact of the “delusions” which we all collectively and individually live under. A great many of them are not based in any kind of objective underlying evidence or “reality” and yet we perpetuate their forms and inequities out of mere habit and call it normal, and imagine that it must have always been so, and so it must always remain.

I think if we’re going to get anywhere in this research authentically, we are going to have to cut a lot closer to the bone.