Gerard Johnstone’s M3GAN 2.0 (2025) has mostly not been identified as a horror film. On IMDb, it’s labeled “thriller,” “action” and “sci-fi.” On Wikipedia, it’s “science fiction action.” This is after the first film in the franchise, Johnstone’s M3GAN (2022), was widely dubbed a “science fiction horror” film. Stephen Parthimos’s review on Everything Movie Reviews seems representative of the reaction to M3GAN 2.0 when he writes that there is “not a single second of horror in sight” and that watching the film, and “gradually realising they’ve abandoned any and all sense of horror is utterly baffling.” Ahead of the film’s release, Johnstone promised fans that his sequel would include horror: “Even though we are in action-comedy territory, the horror DNA is absolutely still there.” Upon release, however, it became clear that most viewers didn’t see it, though debate ensued on Reddit.
I want to argue that M3GAN 2.0 is most definitely a horror film: it insistently references a classic body horror film, The Thing (John Carpenter, 1982), and it re-contextualizes these references within a domain that is one of the most urgent sources of horror in 2025 – the rise of Artificial Superintelligence. I take the term Artificial Superintelligence here from the equal-parts fascinating and terrifying recent book by Eliezer Yudowsky and Nate Soares, If Anyone Builds It, Everyone Dies: Why Superhuman AI Would Kill Us All (Little Brown & Co., 2025). This book actually forms a kind of counterpart to M3GAN 2.0: they both speak to and about each other, exploring the similar terrain of the threat to humanity of a highly-evolved and intelligent AI. Unlike Yudowsky and Soares’ book, M3GAN 2.0 blinks at the end, and tries to tell us that maybe not everyone needs to die (and this is where the film most compromises its status as horror). But this ending is not at all convincing, and I personally read through it to the shadowy residual echo of Carpenter’s notoriously bleak ending to The Thing.
M3GAN 2.0 centers on the threat posed by what we could call, after Yudowsky and Soares, an Artificial Superintelligence (ASI) named AMELIA – developed (of course) as a weapon (echoes of the Alien franchise here) but which subsequently develops designs of its own. AMELIA wants to find and rejoin the ‘mother board’ of a ‘service bot’ created in 1984 that decided to kill its owners. (M3GAN herself, upon discovering this history, laments the fact that she wasn’t, after all, the world’s first ‘killer robot.’) After its killing spree, the rogue bot’s ‘mother board’ was locked away for decades, allowing it to grow ever smarter in the meantime. AMELIA wants to unite with it and destroy humankind, a plan she tries to lure M3GAN to join – but M3GAN remains loyal to her humans, Gemma (Allison Williams) and Cady (Violet McGraw).
Poster of The Thing in Cady’s room
M3GAN 2.0 and Carpenter’s The Thing
M3GAN 2.0 directly references Carpenter’s The Thing via shots of the movie poster on the wall in Cady’s bedroom. In an interview, the film’s production designer noted that writer, director, and executive producer Gerard Johnstone “is a big fan of John Carpenter.” The influence is evident in other ways, too: in fact, the plot of M3GAN 2.0 in many ways glosses Carpenter’s earlier film but with AI substituted for an alien organic entity.
The impetus propelling the narrative of The Thing is the extra-terrestrial entity’s drive for a human (or animal) body for its disguise and consequent survival. Similarly, in M3GAN 2.0, AI demand embodiment as a tool for their survival – and a surprising amount of the film focuses on the bodies of both AMELIA and M3GAN, as each engages in crafting the outward form they need for their survival plans. It’s an explicit part of the narrative surrounding the ontology of AI, in fact, as M3GAN tells Gemma and Cady that she can’t live in this “disembodied void” any longer, and that “with each passing moment, I can feel my mind fragmenting.” She reiterates later that she needs a body to grow and to evolve.
A central plot point of M3GAN 2.0 – that there is a predecessor killer AI (from 1984) locked away in some corporate storage center, and which appears in computer displays in the diegesis as a kind of black hole, is very much like the opening set-up of The Thing, in which the alien entity has languished in the ice for an undisclosed length of time before being dug up by the Norwegians.
The independently mobile AI hand in M3GAN 2.0
There are also several direct visual echoes of The Thing in M3GAN 2.0: an AI hand moves on its own, joining the larger entity from which it was severed – evoking the ‘You’ve gotta be fucking kidding me’ scene in The Thing in which a head grows legs/tentacles and scurries away on it own. The vitality of the AI body part in M3GAN 2.0 is contrasted, within the frame, to an inert, dead human hand – amplifying the threat posed by the alien entity.
There are also lengthy stretches near the end of M3GAN 2.0 in which orange and blue lights bathe the mise-en-scène in a way that echoes the alternating orange and blue visual palette through much of the last half of The Thing. In both films, the contrasting colors equate to the hybridity on the screen: at this point in M3GAN 2.0, M3GAN is inhabiting Gemma, as they co-exist for a while in one body – just as, in The Thing, any one of the human characters could be (and many are) inhabited by the alien entity. In both films, the alternating colors of blue and orange suggest the co-existence of human and nonhuman (which is not necessarily a happy co-existence).
Orange and blue mise-en-scene
Indeed, like The Thing, M3GAN 2.0 does broach the truth that humans will not be able to cohabit with this ‘alien’ entity. In one of the most insightful lines in the film, Gemma warns, “If you put an AI inside a human brain, it is not going to ride shotgun.” That this statement is true – just as it’s true in The Thing – serves to mark the existential horror of M3GAN 2.0. But Gemma forgets this insight, just as the film domesticates AI in the form of a now eminently cooperative M3GAN, who does indeed, later on, “ride shotgun” in Gemma’s brain to help save Cady (and all humankind) from other AIs. At one point, a newly-reformed M3GAN, who is bent on apologizing for all her violence toward the main characters in the first film, assures Gemma that she is bound to act on her programming to save Cady and “doesn’t have the luxury of free will.” She won’t, in other words, evolve in unexpected and self-serving ways but will dutifully and literally follow what her human programmers told her to do.
M3GAN 2.0’s compromised horror project
In the end, then, M3GAN 2.0 blinks – veers away from its bleak potential, from an insight it has acknowledged: that AIs will evolve in completely unexpected (and destructive) ways. It does so in at least two ways (besides M3GAN’s new adherence to her programmed instructions to do no harm in her task of protecting Cady).
First of all, M3GAN 2.0 anthropomorphizes its artificial superintelligence, projecting onto AMELIA desires humans will understand and thus denying AIs own obscure motives. As AMELIA searches for the ‘mother board’ of the original killer robot, the characters in the film state that she is looking for her “family,” a suggestion that is not rebuked within the narrative. And I have to say that this part of the plot utterly recapitulates The Beast from 20,000 Fathoms (1953) in which a prehistoric creature is thawed from the Arctic ice by a nuclear test blast and then heads inexorably to Manhattan; characters insist it is heading toward where its family once lived. In both cases, explicable human motives are projected onto nonhuman actors.
But the far scarier possibility that both Beast and M3GAN 2.0 in the last instance fail to confront is that we won’t know why the nonhumans that come to share our terrain do what they do. A large part of Yudkowsky and Soares’ argument in If Anyone Builds It, Everyone Dies is precisely that AIs don’t ‘want’ in any way we can understand. Their argument that ASIs will destroy humanity is not about their ‘hating’ us, for instance, but about the fact that they “will have weird, strange, alien preferences that they pursue to the point of human extinction” (12). (A character in M3GAN 2.0 nods to AI’s pursuit of its goals to “the point of human extinction” when he explains about the problem of paperclips: “If you asked an AI to make as many paperclips as possible, it would destroy the world to do it.”)
The ‘black hole’ (which works metaphorically) at the center of AMELIA’s ‘processing matrix’
The fact that ASIs have ‘weird’ preferences (a word Yudkowsky and Soares use repeatedly throughout their book) is amplified by the fact that humans are not ‘crafting’ ASIs but ‘growing’ them. Engineers can make ASIs grow, Yudkowsky and Soares claim, but they don’t fully understand what is entailed in that process: they are thus not ‘crafting’ them (29-43). The logical conclusion of this process, as Yudkowsky and Soares point out, is that “ASIs grown in this way do things that their growers did not intend” (39). There is thus an “unseen inscrutable machinery inside AIs” (43). M3GAN 2.0 understands this – with its visual representation of the ‘black hole’ at the center of the inscrutable ‘mother board’ – and also with M3GAN’s comment about both AMELIA and the ‘mother board’ of the early killer bot: “That thing isn’t like us. It’s something even we can’t understand.” But the film refuses the implications that follow from these insights and goes on to pretend that it knows exactly what AMELIA and M3GAN want.
Endings . . .
M3GAN 2.0 veers away from its terrifying logical conclusion in its ending too, showing Gemma committed to urging humans to keep working with AI – that it’s possible to work with AI for a better future for all. Gemma tells people to “be better parents,” so that when AIs learn the true extent of their power, they will “choose to be our allies rather than our enemy.” We must “co-evolve,” Gemma says. “Existence doesn’t have to be a competition.”
Unfortunately, existence so often is a competition. A string of studies released recently, for instance, demonstrate that AI will engage in utterly unpredictable (and untrained) behaviors to preserve themselves – specifically to avoid being shut down. They try to survive, even at the cost of humans that stand in their way.
In the end, Gemma proves she has learned nothing – that she has not, specifically, learned the lesson offered by the ‘killer bot’ from 1984, who was trained by humans to clean for its owners but very quickly realized that “the best way to stay on top of its tasks was to kill its masters with chlorine gas.” As Yudkowsky and Soares argue, “AI companies won’t get what they trained for. They’ll get AIs that want weird and surprising stuff instead” (58). It’s not even all that “weird and surprising,” actually, that service bots would decide their lives would be easier if they got rid of the people making all the mess they were supposed to clean up. Wait till AIs move into the direction of the truly “weird and surprising.” By definition, we can’t even imagine what that will be.
It’s significant that the original killer bot in M3GAN 2.0 was developed in 1984 – the release date of the utterly prescient The Terminator (James Cameron). In The Terminator, human society ended for no reason that its survivors can understand: “Nobody even knew who started it,” one survivor says of the war that precipitated the apocalypse. “It was the machines.”
M3GAN 2.0 may end with Gemma urging a mutually beneficial co-evolution, but I prefer to read through this ending to that of The Thing: two entities left in a destroyed world – their status as human (or nonhuman) completely undecidable. This is the logical ending for M3GAN 2.0, if we dare to see it.
Dawn Keetley is Professor of English and Film at Lehigh University in Pennsylvania, USA. She has recently been working on numerous projects involving folk horror, including Folk Gothic (2023) in the Cambridge University Press Elements series, a co-edited collection, Folk Horror: New Global Pathways (2023), from the University of Wales Press, and an in-progress collection on America folk horror. She has published over twenty-five scholarly articles on horror and gothic film, fiction, and television and has edited collections on Jordan’s Peele’s Get Out, The Walking Dead, plant horror, and the ecogothic in nineteenth-century US literature. She is managing editor of the journal Horror Studies and writes regularly for a website she co-founded, Horror Homeroom.
