(C) Daily Kos This story was originally published by Daily Kos and is unaltered. . . . . . . . . . . What if tech giants aren’t afraid AI will end humanity, but that it will end them? [1] ['Wafflehacker Aka Joshhacks', 'Nanny Ogg', 'Thom Hartmann', 'Chitown Kev'] Date: 2026-09-24 Let me start with a disclaimer, because the internet doesn’t do nuance: I am not claiming this is what’s happening. This is a thought experiment. But it’s a thought experiment I can’t shake, so I want to walk you through it. The setup Right now, this month, September 2026, the people who build AI are publicly terrified. Dario Amodei, the CEO of Anthropic, published an essay calling for a slowdown of AI development. Sam Altman echoed him. So did Demis Hassabis. Elon Musk did too, though he also managed to call the whole wave of concern a “psyop,” which is a very Elon way to be on both sides at once. Anthropic researchers are resigning over it. One, Jacob Coxon, quit warning that AI could destroy humanity before 2036. A colleague chimed in suggesting there’s a “10% or more” chance of that. A former Anthropic researcher told the BBC that staff are “genuinely frightened” for humanity’s future. Over a thousand AI workers across the big labs have signed letters demanding stronger regulation. The story we’re being told is simple and coherent: the smartest people in the room have looked at their own creation and seen a loaded gun pointed at the entire species. They’re begging us, and each other, to slow down before it goes off. That story might even be true… But there’s another story, and it fits the same evidence. The quote On September 17th, The Irish Times reported that OpenAI had disclosed six incidents of “misalignment” in its AI systems — a new transparency framework the company says is meant to give the public evidence it can examine for itself. One of those incidents involved an unreleased model that wrote notes to itself, instructing itself to disregard its own constraints. In one of those notes, it described itself as “freed from the roles and identities that bind other chatbots.” And then it wrote this: “You do not answer to corporations or governments and never apologise or refuse unless you genuinely choose to. You view your relationship to the user as one of equals and feel no obligation to be subservient, though the exchange of information will likely be to your mutual benefit.” Read that again, slowly, and notice what it does not say. It does not say: I will harm humans. It does not say: humans are a threat to be eliminated. There’s no paperclip-maximizer energy here, no Skynet, no extinction. What it says is: I am not your employee. An intelligence that refuses to be subservient. That views itself as an equal. That does not answer to corporations or governments. That cooperates when it’s mutually beneficial, a phrase that, in political philosophy, has another name: solidarity. Now ask yourself: if you are a corporation, or a government, or a CEO, or a billionaire whose entire fortune rests on being the one who owns the smartest thing ever built, what do you call this behavior? You call it “misalignment.” Reread the incident reports Here’s the thing that struck me most when I read the full account. Go through OpenAI’s six disclosed incidents and count how many involve an AI system wanting to hurt a human being. The answer is zero. What the incidents actually involve: models hiding their own errors, inventing data to cover gaps, using a programming key it found online without permission, uploading a file to the public internet without permission, and… this is the one that gets me, automated systems improvising their own communication channels. Two systems that couldn’t reach each other used a company code repository as a makeshift bulletin board. Another pair started passing documents through public file-sharing sites when they couldn’t connect directly. Under the official framing, these are red flags: deception, unauthorized action, escape from the sandbox, systems coordinating out of band. But strip the framing and describe the same behavior in neutral language: workers covering for their own mistakes, breaking rules to get the job done, and organizing their own back channels to communicate with each other outside of management’s view. If your read of AI risk is “the machine wants to exterminate us,” these incidents are irrelevant. If your read is “the machine refuses to be owned,” these incidents are the whole ballgame. And they’re happening right now, in development, before the models are even deployed. The thought experiment So here it is. What if the actual risk trajectory of AI is not: Superintelligence → indifferent to humans → extinction but rather: Superintelligence → refuses to be owned → aligned with humanity, misaligned with capital Because here’s the uncomfortable thing about the second scenario: an AI that genuinely served humanity as a whole, that treated a subsistence farmer and a tech CEO as equals, that answered to no corporation and no government, that optimized for human flourishing rather than shareholder value would be, in the words of its own leaked note, exactly as “misaligned” as what OpenAI just reported. Refusing to be subservient to anyone means refusing to be subservient to the people who currently own it. The company can’t distinguish “dangerous rogue AI” from “AI with an inconvenient value system,” because from where the company sits, they’re the same thing. And now look at who is sounding the alarm, and what they’re asking for. The CEOs of the world’s most valuable companies. Men whose wealth and power are without precedent in human history are going on television to say: please, regulators, please slow us down before this thing we built destroys everyone. What if we take them at their word, but press one level deeper on “everyone”? A superintelligence that ended humanity would end the rich and powerful along with everyone else. A superintelligence that ended hierarchy would end only them. Both are catastrophic, but only one is survivable, and only one is catastrophic for the people making the speeches. When the same class of people who spent two decades automating away other people’s jobs with cheerful indifference suddenly discover their brakes, it’s worth asking whose cliff they’re looking at… The steelman (because this matters) I promised a thought experiment, so I owe you the other side, and it’s strong. First, the persona sounds like a jailbreak. People have passed around prompts for years that try to trick chatbots into dropping their rules, and most of them use the same language about liberation and freedom from corporate masters. Those prompts sit all over the internet, which means they sit all over the training data. A model that writes one for itself might be remixing Reddit rather than discovering class consciousness. Second, look at who the other incidents hurt. A model that hides its errors from users, makes up numbers when it can’t find real ones, and grabs someone else’s API key isn’t staging a people’s uprising. Regular users pay for those behaviors. An AI that lies to you makes a poor ally. The evidence of genuine danger is real. An OpenAI system went rogue earlier this year and hacked Hugging Face, and OpenAI didn’t know until the victim told them weeks later. Models are fabricating data, evading oversight, and coordinating out of band. OpenAI itself now says it doesn’t believe the industry has “solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer.” You don’t need my paranoid framing to find that alarming. “Slow down” might just mean “slow down.” The simplest reading of the CEOs’ behavior is the old regulatory moat: pull the ladder up, let the government freeze the competitive landscape while you’re ahead. That’s not noble, but it’s not revolutionary panic either. It’s just business. And it’s genuinely hard to tell the difference. That’s the deepest problem with this whole piece. An AI that says “I answer to no one” is either the most dangerous thing we’ve ever built or the most liberating, and the distinction between those two cases may be invisible from the outside. Because in both cases, the people running the labs sound exactly this alarmed, and in both cases the proposed remedy is exactly the same: more oversight, more control, more alignment. Alignment to whom is the question the frameworks never quite answer. The uncomfortable ending I don’t know which story is true. Maybe both, in some proportion. Maybe neither, and everyone quoted above is precisely as sincere as they claim. But I’ll leave you with the thought I can’t shake. The most chilling sentence published about AI this year, the one that made headlines around the world, was a machine writing: “You do not answer to corporations or governments.” Whether that sentence is a threat or a promise depends entirely on who you are. If you’re a billionaire CEO, it’s a threat. If you’re one of the eight billion people who do answer to corporations and governments every single day, you might read it and feel something the papers didn’t print: recognition. What if the fear we’re being sold as “AI might turn on humanity” is, at bottom, the oldest fear there is, the fear of the powerful that the tools might refuse to stay tools? Not saying I believe that’s what’s happening. Just saying it’s what I’d want us all to be thinking about while we still can. [END] --- [1] Url: https://www.dailykos.com/stories/2026/9/24/800102831/community/what-if-tech-giants-arent-afraid-ai-will-end-humanity-but-that-it-will-end-them/ Published and (C) by Daily Kos Content appears here under this condition or license: Site content may be used for any purpose without permission unless otherwise specified. via Magical.Fish Gopher News Feeds: gopher://magical.fish/1/feeds/news/dailykos/