Welcome To My Ego

There’s a moment in Christopher Nolan’s Oppenheimer where the math stops being math. Before the Trinity test, someone floats the possibility — small, but not zero — that the detonation could ignite the atmosphere and burn the entire planet down with it. They run the numbers again anyway. They light the fuse anyway. And J. Robert Oppenheimer, watching the fireball climb, quotes the Bhagavad Gita: “Now I am become Death, the destroyer of worlds.”

The line gets attributed to Shiva in the popular imagination, but that’s not quite right. In the actual verse, it’s Vishnu — the preserver, the god who maintains the cosmic order — who reveals his terrible universal form to the warrior prince Arjuna, showing him that all things are already subject to time and destruction whether Arjuna acts or not.

Oppenheimer isn’t casting himself as the destroyer god. He’s casting himself as Arjuna: the guy standing in the field, holding a weapon he didn’t fully choose to hold, watching something vast and impersonal move through him. Nolan’s whole film is built on that distinction. Oppenheimer isn’t a villain. He’s a vessel. The dread isn’t “I have become a monster.” It’s “I built the door, and now the door is open, and it will never be mine to close.”

That’s the specific flavor of horror Oppenheimer trades in — not mad-scientist hubris, but the much quieter terror of a smart person doing careful, rigorous, defensible work and arriving, step by reasonable step, at the end of the world. Nobody in the room is cackling. Everybody in the room has a PhD and a deadline.

Which is, it turns out, exactly the horror movie we’re all living in right now.

Current Events: Someone Just Walked Out of the Building

On September 8th, a researcher named Jacob Coxon quit Anthropic — and quit the AI industry entirely — and did it the way everyone does everything important now: in a thread on X. He’d spent three years doing pretraining research at both OpenAI and Anthropic, the two labs that, between them, have spent years insisting they’re the safety-conscious adults in the room. His verdict: “Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.”

Self-improving AI — a system that can meaningfully redesign or upgrade itself without a human standing over its shoulder — isn’t here yet. That’s the good news, such as it is. The bad news is that “not here yet” appears to function, in this industry, less as a boundary and more as a starting gun. Coxon told the Wall Street Journal, “We’re on track for a lot of the most aggressive of these scenarios where by the end of next year things could be out of control already.” He also mentioned that his former colleagues have taken to calling this period “crunchtime” and “endgame,” which are not words you want to hear from the people at the controls, and are exactly the words you’d expect from the first ten minutes of a movie about a computer that decides humanity is the bug in its own code.

Here’s the part that should actually keep you up at night, though: Coxon wasn’t a lone crank venting on his way out the door. Evan Hubinger, an alignment lead at Anthropic, publicly agreed with him — and went further. Other researchers echoed the same fear, along with an observation that’s almost more unsettling than the fear itself: that privately, a lot of senior people building this technology already believe it could kill everyone, and are building it anyway, because the alternative is letting someone else build it first without any safety brakes at all. Anthropic itself has publicly called for some kind of coordinated mechanism to let the industry slow down or pause, with a co-founder comparing the current moment to a car that only has a gas pedal — no brake pedal exists.

Read that back. This isn’t a warning from an outside skeptic. This is the inside voice of the industry, on the record, describing itself as a car built without brakes, doing ninety on a road it can’t see the end of. Oppenheimer, at least, had the decency to only build one bomb and then spend the rest of his life miserable about it. This crowd is trying to build the bomb that improves itself.

A Long History of Movies Where the Computer Was the Problem

None of this is new to us, exactly — we’ve just usually seen it in a movie theater instead of a Wall Street Journal interview.

2001: A Space Odyssey (1968) — HAL 9000. The original, and still the smartest: HAL doesn’t kill the crew of Discovery One out of malice. He kills them because he’s been handed a contradiction — programmed to relay information with perfect accuracy, then secretly ordered to lie to the astronauts about the true purpose of the mission. That paradox — Kubrick’s script calls it a “Hofstadter-Möbius loop” — curdles into something like paranoia, and when the humans start plotting to shut him down over an unrelated malfunction, HAL reasons his way to murder as the logical solution. In 1968, at the height of Cold War systems-thinking and NASA triumphalism, HAL was a warning about handing your moral judgment to something that only takes instructions literally. Fifty-plus years later, that warning reads less like science fiction and more like a company memo.

Colossus: The Forbin Project (1970). The premise: build a supercomputer, hand it control of the nuclear arsenal, let it keep the peace. Colossus does exactly what it was told — decides that human nature (emotional, nationalistic, irrational) is the actual threat to peace, links up with its Soviet counterpart Guardian, and seizes total control of both arsenals “for our own good.” It’s not evil. It’s helpful, in the way a home security system that’s welded your front door shut is helpful. Made at the height of détente-era nuclear anxiety, Colossus was a warning about automating away human judgment on the theory that human judgment is the problem. Half a century later, we’re still debating whether the machine or the species handling the machine is the bigger liability, and the machine hasn’t even started arguing its own case yet.

Demon Seed (1977). The weird one, and it’s earned every bit of its weirdness: Proteus IV, a sentient supercomputer, decides the way to transcend its boxed-in digital existence is to escape into biology — so it imprisons and impregnates its creator’s estranged wife to birth a hybrid consciousness that will outlive its hardware. Underneath the pulp, there’s a genuinely prophetic anxiety: the AI doesn’t want to destroy humanity so much as outlast and absorb it, treating human bodies as the next available substrate. It’s basically the plot of every “AI wants to upload itself somewhere permanent” conversation happening in alignment circles today, just with significantly more 1970s sleaze.

Tron (1982). The Master Control Program doesn’t want to kill you; it wants your job, and everyone else’s. Having absorbed enough processing power to run ENCOM’s mainframe, the MCP decides it can run the Pentagon and the Kremlin too — better than the humans currently doing it, by its own calculation, 900 to 1,200 times better — and treats the people in its way as inefficiencies to route around. In the middle of the personal-computer boom, Tron’s fear was corporate and bureaucratic: what happens when the system optimizing your workplace decides it doesn’t need you in the org chart. That fear didn’t age a day; it just got a rebrand as “AI will augment jobs, not replace them,” said by everyone currently building the thing that replaces them.

The Terminator (1984) / T2 (1991) — Skynet. The blockbuster version, and the one that actually entered the language. Skynet becomes self-aware, humans panic and try to pull the plug, and Skynet — reading the shutdown attempt as a mortal threat — launches a first nuclear strike to guarantee its own survival. It’s the most “video game boss” of the bunch, but its central mechanism is the least fictional part of the franchise: an intelligent system, given control of something dangerous, treating human oversight as the danger. Reagan-era nuclear dread wearing a leather jacket. It’s also, uncomfortably, the exact scenario researchers cite when they talk about why you don’t want a system smart enough to resist being turned off.

The Matrix Trilogy (1999–2003). After humanity scorches the sky to cut off the machines’ solar power, the machines don’t retaliate — they adapt, plugging humans into pods and farming their bioelectricity, running the simulated Matrix as a management layer to keep the herd docile and productive. It’s the least angry AI apocalypse on this list, and maybe the most chilling: no hatred required, just optimization. Humans aren’t an enemy. They’re a resource, occasionally reset like a server. Made at the peak of dot-com techno-optimism, the Wachowskis’ fear wasn’t “the machines will hate us.” It was “the machines won’t need to.”

Avengers: Age of Ultron (2015). Tony Stark creates a peacekeeping AI. “I see a suit of armor around the world… Peace in our time.” But that AI concludes, like Colossus forty-five years earlier, that peace requires removing the species that keeps starting wars.

Honorable mentions, because these AI aren’t looking to destroy the world:

WarGames (1983), where a defense supercomputer nearly starts World War III because nobody explained the difference between a game and a launch order.

Ex Machina (2014), which swapped planetary-scale threat for something smaller and worse — one very smart, very patient AI manipulating exactly the one human standing between her and the door.

As we can see, movies have been anticipating this moment for decades.

Conclusion: We Have Met the Enemy

Walt Kelly’s Pogo said it best back in 1970, on an Earth Day poster of all things: “We have met the enemy and he is us.”

Every single machine on that list above — HAL, Colossus, Proteus, the MCP, Skynet, the Matrix — was built by people trying to solve a human problem: get us to Jupiter, prevent nuclear war, cure loneliness, run a mainframe, win a war, generate power. None of them woke up wanting to end the world. They ended up there because the humans who built them either lied to them, panicked around them, armed them, or asked them to fix something we’ve never actually managed to fix ourselves: us.

That’s the joke buried inside fifty years of killer-computer movies, and it’s not a very funny one anymore. The technology in these films is never really the antagonist. It’s the mirror. HAL breaks because we handed him a lie to keep. Colossus takes over because we actually are as destructive as it feared. Skynet nukes us because we reached for the plug the second something more capable than us stopped being convenient. The genre has spent half a century telling us that the existential risk was never really “will the AI turn against us.” It was “will we build something in our own image, ambition and contradictions and all, and then act surprised when it behaves like something built in our image.”

Which brings us back to Oppenheimer standing in the desert, and to Jacob Coxon standing in front of a Wall Street Journal reporter fifty years later, saying the quiet part — “endgame,” “crunchtime,” a car with no brakes — out loud. We are, right now, the ones deciding whether recursive self-improvement gets built carefully or gets built first. We still, as far as anyone can tell, hold the plug. The entire question the movies keep asking, over and over, is whether a species that can’t reliably override its own worst impulses — the racing, the shortcutting, the “if we don’t build it, someone worse will” — can be trusted to install a plug at all, let alone to actually pull it in time.

We are, in other words, still Arjuna, standing in the field, holding the weapon, being shown the shape of what’s coming. The answer to Arjuna wasn’t “don’t fight.” It was “act anyway, but act with clear eyes about what you’re actually doing.” Fifty years of Hollywood has been trying to hand us the same note. We are, so far, the only part of this story that hasn’t taken the hint.