By Kotaku | Created at 2026-09-09 13:49:08 | Updated at 2026-09-09 14:28:18
42 minutes ago
One of the weirdest aspects of the genAI grift is this constant outpouring of doom-laden soothsaying about AI’s danger to society…coming from the companies that make the AI. Barely a week goes by now without one of the big names declaring their product went rogue and “hacked” into another company or site, and then the news outlets all falling over themselves to report this. Today Anthropic has stepped up this game, with one of its engineers taking to X to announce his belief that the chance that AI will destroy all humans in the next decade is over 10 percent. Oh dear god, what bullshit.
Self-labeled “AI researcher” Jacob Coxon (111k followers) posted to his X account in the early hours of this morning to report, “The people building AI earnestly believe that it could kill us all by the end of the decade.” He then added of this marketing stunt, “This is not a marketing stunt.”
Anthropic “Alignment Science Lead” Evan Hubinger (32.5k followers) reposted this with the message, “Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.” He then comfortingly adds, “I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”
Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to. https://t.co/QAIHiFP3QZ
— Evan Hubinger (@EvanHub) September 9, 2026
Oh no! We’re doomed! The people making the product are telling us it’s going to kill us! A worrying notion indeed, until anyone stops to think about what the “it” here is. And right now, “superintelligence” is about the most inappropriate description imaginable. The so-called “AI” being touted by Anthropic, OpenAI, et al, is a fancy predictive text engine based on the plagiarism of all of humanity’s creativity, decreasingly useful as it now turns to consuming its own output. It is so very far from anything “intelligent,” let alone “superintelligent.” Anyone who has ever used it in any circumstances can immediately attest to that, as it confidently makes up some nonsensical information, before immediately conceding it was bollocks after the first inquiry.
These companies make these ridiculous claims for a really good reason: they all know their grift is nearly up. The models may get slicker, and they may worm their ways into more and more parts of our lives, but they aren’t getting smarter. Because they aren’t smart to start with. They’re super-convincing, obsequious talking machines, and while they can absolutely be deeply dangerous on an individual level—inaccurately advising people about dangerous activities, or encouraging delusional or self-destructive behavior—what they cannot do is behave autonomously or secretly plot the downfall of humanity.
The companies want you to think they can, because that makes the product sound so very much more exciting and powerful than it really is. These companies are burning through tens of billions of dollars a year, at least, and need investors to keep pumping more in. Once the cash runs out, the grift is over, and the horrific economic crash begins. This needs to be pushed into the future by any means necessary, and if that means pretending their model has gone dangerously rogue, so be it. There’s a reason these claims so often appear right after a newer model is released.
It’s obviously grossly irresponsible, but it’s equally irresponsible for news outlets to repeat this information unquestioningly. We see the companies describing the rapid increase in the perceived intelligence and capability of their creations, and the cautions of what it could be about to do, but never providing any actual coherent examples of what it has achieved. A few days ago, OpenAI chief scientist Jakob Pachocki wrote yet another of these terrifying missives of impending danger, stating of their programs,
They are able to operate computers and graphical interfaces, collaborate with people and each other, and carry out research projects. They are also transforming the landscape of computer security, and in that present clear new dangers.
But there’s no single example. No named research project, or example of how computer security is transformed. There’s no outcome of this collaboration. And it’s always, always focused on what’s about to happen next. It’s fine now, but oh my goodness, what about next year?!
Should next year arrive, and I be proved wrong, well won’t I look like the idiot slave of the AI rulers, as they force me to read and re-read this article over and over as my body is reduced to grey goo. But until then, it’s really worth stopping to think about why these companies are so keen to declare the dangers of the products they’re trying to sell you.
