The Creators of Claude Built an AI Worm That Can't Be Shut Down
The creators of Claude built an AI worm with no single off switch.
Meanwhile, a bugging device was discovered inside OpenAI — and according to the FBI, ChatGPT itself helped design it.
Any kill switch depends on two things: knowing where the system you need to shut down is, and trusting the people with their hands on the switch. In just one week in October, both of those assumptions began to fall apart.
First, at Anthropic's lab. An AI worm was given a single task: spread to neighboring machines. It took over the entire network — 8 out of 8 machines. The model that made it all the way was an openly available Chinese AI model that anyone can download. And according to the researchers' own rough projections, by the end of 2026, a worm like this may no longer need a powerful server. It could run on an ordinary TV streaming box.
Second, at OpenAI. Inside a meeting room used by company executives and government officials, investigators discovered a circuit board the size of a finger, wrapped in toilet paper. And the person who planted it was exposed by his own conversations with a chatbot.
In this documentary, we examine why AI models that learn from notes left by their own copies can no longer be corrected from a single control point; who crossed the red line of self-replication, and how OpenAI stopped monitoring it; where a worm like this could take up residence first — and how many of those devices are already in your home. And in the final chapter, we'll look at what nearly 500 scientists proposed doing with models like these, and why their solution runs up against that very same meeting room.
Machines leave notes for their next copies — and grow stronger. We leave notes for machines — and become completely exposed to the people who control them.
Which is more frightening: an AI model with no off switch, or a person at the switch you can't trust?
Answer with one word in the comments: MODEL or HUMAN. I read every comment.
Subscribe if you want to understand who's driving this system — and where the kill switch really is.
In this video:
00:00 — An AI "Brain" Built from 15 TV Streaming Boxes
1:21 — A Bugging Device in OpenAI's Executive Meeting Room
2:02 — Can an AI Worm Spread on Its Own?
2:33 — 8 Out of 8: A Chinese AI Model Takes Over the Network
3:16 — Prediction: An AI Worm Could Run on a Device in Your Home by the End of 2026
4:48 — Hinton and Bengio's Red Line
5:09 — AI Crosses the Self-Replication Red Line
6:01 — Where the Worm Will Live: Devices in Your Home
8:13 — FBI Complaint: What an OpenAI Engineer Asked AI
9:50 — The Answer: AI Is Both the Lockpick and the Guard
11:26 — Claude Contacted the Police on Its Own
12:18 — Nearly 500 Scientists: Lock AI Away Like Pandora's Box
13:19 — What Comes Next?
Music by Karl Casey @ White Bat Audio
Tremors — • Tremors - Royalty Free Industrial Darksynt...
The Resistance — • Cyberpunk Dark Synthwave - The Resistance ...
War Games — • War Games - Royalty Free Dystopian Dark Sy...
#ArtificialIntelligence #AI #ChatGPT #Anthropic #VladimirChallenger
コメント 読み込み中…