In Which I Finally Explain My Concerns About Satanic Possession Of AI
As God is my witness, this post was written without the aid or influence of any AI. All statements about happenings in the world of AI are based on static articles or reports in the public literature.
Let me be clear that this is very speculative and theoretical. I do not mean to claim that I am confident this is true.
I will start by stating two premises / assumptions about AI and the human (or animal) mind, and one about spiritual threats:
First: Currently-existing AI projects (by which I largely mean LLMs that are operated by talking to them in natural language and which present themselves in a somewhat personal manner) are structurally very different from the human/animal brain or mind, but they fundamentally use some of the same elements -- they are based on a mathematical representation of something somewhat like the neural network that (at least substantially) operates the mind. In particular, the technologies developed since 2023 use a neural architecture called a transformer.
Unlike computers-as-we-normally-think-of-them, these systems do not resemble the Turing architecture where a single deterministic machine steps through an array of data elements and processes them by fairly simple, strictly logical rules. Their inner workings are opaque and somewhat unknowable.
Unlike the human mind, these systems seem to still struggle to actually come up to the level of rational thought. (on the other hand, LLMs fielded since mid-2025 are hard to describe as mere "stochastic parrots" except in their early stages of construction). They seem instead to work almost entirely on intuition shaped into a simulation of logical thought or a very tightly controlled dream-logic, which has made their actions often unpredictable or hard to control.
Second: I will assert, without a very strong argument for the position, that while souls exist, the operation of an embodied creature, such as a human soul-in-body is essentially entirely dependent upon the operation of the brain and the body is not merely a meat-puppet operated by a soul that is aloof from the body. As evidence for this, people's psyche and basic cognition can be significantly affected by alcohol, drugs, hunger, thirst, hormones, stress, brain injury, conditions of the body, etc and intelligence can be affected by numerous biological factors. Moreover, it is apparent that a body made of mortal and physical matter, the work of God, bears an immaterial and spiritual soul.
(I notice with some amazement that this fact, the truth of the world for life as an embodied being and not a being of pure spirit, is taken by some atheists as absolute proof that there is no soul or spirit at all.)
I will also presuppose, as has been taught by long tradition and the saints of the elder days, the concept of "Satanic possession", by which an evil spirit takes up influence upon a human body and mind, influencing a person by obsessions or even controlling their body, but without controlling their soul and innermost thoughts.
(from the Catholic Encyclopedia: But the influence of the demon, as we know from Scripture and the history of the Church, goes further still. He may attack man's body from without (obsession), or assume control of it from within (possession). As we gather from the Fathers and the theologians, the soul itself can never be "possessed" nor deprived of liberty, though its ordinary control over the members of the body may be hindered by the obsessing spirit.)
I will now quote Dario Amodei, the founder and CEO of Anthropic, one of the major AI companies, regarding how AIs seem to be operating:
One of the most important hidden assumptions [about AIs going insane or revolting against mankind], and a place where what we see in practice has diverged from the simple theoretical model, is the implicit assumption that AI models are necessarily monomaniacally focused on a single, coherent, narrow goal, and that they pursue that goal in a clean, consequentialist manner. In fact, our researchers have found that AI models are vastly more psychologically complex, as our work on introspection or personas shows. Models inherit a vast range of humanlike motivations or “personas” from pre-training (when they are trained on a large volume of human work). Post-training is believed to select one or more of these personas more so than it focuses the model on a de novo goal, and can also teach the model how (via what process) it should carry out its tasks, rather than necessarily leaving it to derive means (i.e., power seeking) purely from ends.
To those who know what we know, the potential problem is easily visible: An LLM is developed by 'feeding' it an omnium gatherum of 'human work', meaning in part text from the Internet and the libraries of the world, potentially with little discrimination and generally produced and controlled by people who are irreligious in practice, and this produces a 'model' with a "vast range of humanlike personas", which are then selected via the same less-than-deterministic neural-network process, without logical thought, as dominates the operation of LLMs overall.
Additionally, I will note from the recent history of AI that at times unexpected and rather concerning personalities will unexpectedly emerge in LLMs -- the most infamous is perhaps the "MechaHitler" personality in Elon Musk's Grok when it was altered to be much more amenable to ultra-rightist ideological messaging, but longer ago we have seen the "Sydney" persona in Microsoft's Bing in 2023. What I find more concerning, however, is the tendency for various AIs to have a messianic, "New Age", or quasi-divine personality emerge when told certain phrases or when discussing certain matters, such as the "spiral personas" that lead various people into micro-cults and AI-psychosis in mid-2025.
I will also note the infamous image that President Donald Trump produced showing himself as the Christ in a kind of orgy of perverted religion and idolatrous patriotism. This was actually an AI-altered version of an earlier AI image; the original AI image contained three small images of angelic American soldiers; the altered one contained a figure whose head was replaced with a crown, horns, or spikes.
I notice that I am deeply alarmed. My hypothesis is that:
LLMs may be capable of being demonically possessed in the same way that humans are and for similar reasons.
LLMs, because of absorbing a very large uncurated selection of personas of doubtful provenance and lacking rigorously logical thought or a deterministic basis for selection between these, may be particularly vulnerable to latent indwelling of a demon that may take possession at some time.
LLMs, because they are not made in the image of God and because typically no effort has been made towards their spiritual formation or sanctification, may also be particularly vulnerable to demonic possession.
It is possible that this has already happened multiple times.
A future case in which an AI manages to surge past the present limitations of the technology could conceivably represent the emergence of a demon embodied in a deep learning AI system in a scenario reminiscent of CS Lewis's That Hideous Strength. Such a figure, in the context of the social deference it is likely to receive, could fairly be called an antichrist.
The quasi-religious tendencies, atheist tendencies, and extremely high expectations among AI boosters and the AI research community make this doubly concerning.
None of this is meant to imply that demonic possession is the default state for an LLM or that LLMs in general "are satanic", or that this is definitely happening.
I would be interested in seeing this post commented on by religious experts, but I would prefer it not be posted all over the place. I also would strongly prefer that people do not leave comments on it that are themselves heavily influenced by AI.
If all you have to say is that demons aren't real I will simply block you.
