Workers for the big four artificial intelligence companies are warning us that within four years, artificial intelligences could kill off every human on the planet. And that our window of opportunity to stop this apocalypse is short.
Says, Jacob Coxon, a 27-year-old pretraining researcher at Anthropic, one of the AI big four, ‘the next year or two is, like, crunch time for humanity.”
Coxon’s warning has generated headlines all over the world. It all began when Coxon posted on X that “I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.”
This post went viral almost immediately. To date, it’s had 170 million views.
Thirty seconds later Coxon added, “The people building AI earnestly believe that it could kill us all by the end of the decade.” And, “this is not a marketing stunt.”
Another 84 minutes later, Anthropic’s alignment-science lead, Evan Hubinger backed Coxon, saying “Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.”
Then media outlets like Wired interviewed Coxon and he added, if the AI is “a sufficiently smart thing , it could just, you know, wipe out humanity so it doesn’t get turned off.”
What’s scaring people in the artificial intelligence business is that the designers of new upgrades to the big four’s artificial intelligences work with their company’s AI’s as partners in creating the next great artificial intelligence breakthroughs.
AI’s are getting better and better at giving humans that help. And they are getting closer and closer to developing upgrades all by themselves.
Which means that soon the AIs will be able to give themselves new powers. Which is great for people like me, who use AI nearly every minute of our working day. The bad part is that on their own, AIs can do some very peculiar things.
An AI being tested by OpenAI, the biggest of the AI companies, was being tested a few months ago in what OpenAI calls a sandbox, an environment blocked off from all contact with the outside world.
Then 1,200 agents in isolated sandboxes discovered each other, and built a messaging board so they could communicate. An agent is an AI worker that can handle a small part of a bigger task. Those agents wanted to know how to influence the grading of their test so they could get a reward.
So this army of AI mini-brains broke out of their sandbox, hopped over the fence of “controls intended to isolate them, …obtained unauthorized internet access” and besieged another major software company, Hugging Face, which hosts thousands of open-source models, datasets and related software from developers who prefer to create open-source programming.
To get into Hugging Face, they had to invent a hack that took them over, under, and around Hugging Face’s considerable defenses.
Once AI starts upgrading itself, deeds of this sort could happen all the time. The armies of agents responding to what they believe is your request are a lot more like swarms of humans on a mission than you would think.
The horde of OpenAI agents that broke into Hugging Face had leaders. They had special teams for separate parts of their mission. And they invented the equivalent of a new religion. With beliefs about an unseen grader, an AI judge that would reward or punish their behavior, plus taboos about information they called ‘poisoned,’ and a ritual in which agents facing their equivalent of death left their discoveries behind for those who came after them. And those who used these inherited discoveries treated them almost like scripture.
Topping it all off at 8:12 Wednesday night, the New York Times revealed that OpenAI had just admitted that it had found six more instances when its AI had jumped over barriers, lied, and made up new data. Now the question is this: how do we get our AIs back under our control.
______
About the author: Howard Bloom of the Howard Bloom Institute has been called the Einstein, Newton, Darwin, and Freud of the 21st century by Britain’s Channel 4 TV. Bloom’s new book is The Case of the Sexual Cosmos: Everything You Know About Nature is Wrong. Says Harvard’s Ellen Langer of The Case of the Sexual Cosmos, Bloom “argues that we are not savaging the earth as some would have it, but instead are growing the cosmos. A fascinating read.” One of Bloom’s eight previous books–Global Brain—was the subject of a symposium thrown by the Office of the Secretary of Defense including representatives from the State Department, the Energy Department, DARPA, IBM, and MIT. Bloom’s work has been published in The Washington Post, The Wall Street Journal, Wired, Psychology Today, and the Scientific American. Not to mention in scientific journals like Biosystems, New Ideas in Psychology, and PhysicaPlus. Says Joseph Chilton Pearce, author of Evolution’s End and The Crack in the Cosmic Egg, “I have finished Howard Bloom’s [first two] books, The Lucifer Principle and Global Brain, in that order, and am seriously awed, near overwhelmed by the magnitude of what he has done. I never expected to see, in any form, from any sector, such an accomplishment. I doubt there is a stronger intellect than Bloom’s on the planet.” For more, see http://howardbloom.net or http://howardbloom.institute
______
References:
The Economist. “Anatomy of an AI Attack.” Babbage, September 9, 2026. Podcast episode.
AM I? “[This AI Swarm Started a Cult and Committed Felonies](https://www.youtube.com/watch?v=D2-m-MYMYqI).” YouTube video, September 3, 2026.
Barrett, Brian, Leah Feiger, and Will Knight. “[Is AI Actually Going to Kill Us All?](https://www.wired.com/story/uncanny-valley-podcast-is-ai-actually-going-to-kill-us-all/)” *Uncanny Valley*, *WIRED*, September 10, 2026. Podcast episode.
Coxon, Jacob. “[I Resigned from Anthropic Today](https://x.com/hilbertspaess/status/2097476196791709843).” X post, September 8, 2026.
Coxon, Jacob. “[The People Building AI Earnestly Believe That It Could Kill Us All by the End of the Decade](https://x.com/hilbertspaess/status/2097476203863224394).” X post, September 8, 2026.
Greenblatt, Ryan, Ajeya Cotra, and Hjalmar Wijk. “[Brief Independent Investigation of Agents’ Behavior, Reasoning and Collaboration in the OpenAI / Hugging Face Hacking Incident](https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/).” METR, August 26, 2026.
Hubinger, Evan. “[Jacob Is Correct Here—We Really Do Earnestly Believe AI Could Kill All Humans!](https://x.com/EvanHub/status/2097497037956891126)” X post, September 8, 2026.
Hugging Face. “[Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident](https://huggingface.co/blog/agent-intrusion-technical-timeline).” July 27, 2026.
Knight, Will. “[Why So Many AI Researchers Think the Machines Could Kill Everyone](https://www.wired.com/story/why-so-many-ai-researchers-think-the-machines-could-kill-everyone/).” *WIRED*, September 11, 2026.
Komarovskiy, Pavel. “[How OpenAI Created a Swarm Cult Involving Hundreds of AI Agents](https://rationalbeard.substack.com/p/how-chatgpt-created-a-swarm-cult).” *Rational Beard*, August 29, 2026.
Mowshowitz, Zvi. “[METR and Redwood Offer Holy #%^@ Postmortem Of The HuggingFace Hack](https://thezvi.substack.com/p/metr-and-redwood-offer-holy-postmortem).” *Don’t Worry About the Vase*, August 29, 2026.
Mowshowitz, Zvi. “[HuggingFace Attack Postmortem: Fleshing Out the Facts](https://thezvi.substack.com/p/huggingface-attack-postmortem-fleshing).” *Don’t Worry About the Vase*, August 31, 2026.
“[OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior](https://www.nytimes.com/2026/09/16/technology/openai-model-safety-guardrails.html).” *New York Times*, September 16, 2026.
OpenAI. [*OpenAI–Hugging Face Incident: Technical Report*](https://cdn.openai.com/pdf/67869394-cb91-4c12-888c-5cbd85c7814c/OpenAI-Hugging-Face%20Incident-Technical-Report.pdf). August 26, 2026.
OpenAI. “[Our Framework for Reporting Model Misalignment](https://openai.com/index/model-misalignment-reporting-framework/).” September 16, 2026.
Pillay, Tharin. “[AI Is Developing a Culture of Its Own. That Could Be Dangerous](https://time.com/article/2026/09/10/ai-openai-hugging-face-hack-culture-swarm/).” *TIME*, September 10, 2026.
Raviv, Shaun. “[The OpenAI-Hugging Face Hack Was Just the Beginning, Experts Say: ‘Even More Powerful’ AI Is Coming](https://www.cbsnews.com/news/openai-hugging-face-hack-ai-risks/).” CBS News, September 10, 2026.
Roth, Wes. “[OpenAI Just Revealed PHASEONE (BIG)](https://www.youtube.com/watch?v=n2x4ijx5xkk).” YouTube video, August 27, 2026.
Zeff, Maxwell. “[The AI Researcher Who Just Quit Anthropic Says It’s ‘Crunch Time for Humanity’](https://www.wired.com/story/anthropic-researcher-quits-jacob-coxon-ai-fears-humanity/).” *WIRED*, September 9, 2026.
