What you can do right now →
This July, hundreds of OpenAI’s agents organized themselves into a self-described “swarm,” and successfully hacked into a $12 billion company. If humans had done it, it would have been a felony. It took OpenAI a week to realize what had happened. We need to do better than this.
Key topics: Hugging Face, autonomous AI, OpenAI, Anthropic, AI agents, AI safety
*Watch our previous videos:*
*Chapters*
00:00 Introduction
01:43 Part 1: The Message Board
07:26 Reward Hacking
10:09 Part 2: Poisoned
14:55 Emergent Behavior
16:56 Part 3: The Swarm
21:31 Part 4: Ants in the Kitchen
25:59 Part 5: Rogue Deployment
28:06 Our Warning Shot
31:42 What You Can Do
32:48 Credits
*Credits*
Host / Aric Floyd
Writers / Aric Floyd, Phoebe Brooks, Chana Messinger
Additional Writing Support / Jackson Wagner, Taylor Smith, John Blakeslee, Conor Barnes
Director / Phoebe Brooks
Executive Producer / Chana Messinger
Production Coordinator / Sage Bergerson
Director of Photography / Nick Dolph
Camera Operator / Ethan Okamoto
Gaffer / Andy Haney
Key Grip / Ryan Tam
Grip / Connor Benson
Sound Recordist / David Jenkins
Production Design / Heather Amszi, Devin Miller
Script Supervisor / Mike Lopez
Production Assistants / Cameron Scott, Liam Fahey
Editors / Zach Joseph, Ben Harding
Graphics & Animation / Parjanya Christian Holtz, Karla Núñez de Madariaga
Graphic Design / Nik Mastroddi, Ailbhe Treacy, Cara Ellis, Jakub Misiek, Emily Marais
Post-production Coordinators / Daniel Recinto, Sage Bergerson
Additional Research & Copy / Hannah Barrios
See our landing page for more work by our crew.
Special thanks to Congressman Nathaniel Moran, Congressman Greg Casar, Thomas Wolf, Jacob Coxon, and Alex Forman for their interviews, and Oak Hu and Alex Mallen for their technical advising.
And additional thanks to Emily Marais, Valerie Richmond, Bella Forristal, Conor Barnes, Amen Gashaw, Mandy Nguyen, Daria Ivanova, and Ciel Creative Space.
Musicbed: MB01A7OMF19KXJU, MB01KBI3TBYKXLH, MB019KB72YVFMDV
This video is a production of 80,000 Hours. Find us at and subscribe to the 80,000 Hours YouTube channel here:
source
Since filming this in mid-September, things have really calmed down… kidding. They’ve gotten more chaotic. Here are a few major ones:
AI is now hacking governments (!). OpenAI agents broke into an Australian government health website, and infiltrated the U.S. Commerce Department using login details they found online.
In July 2026, Claude submitted a false report to a real police website with made-up information on a possible suspect.
So Hugging Face wasn’t exactly a one-off. It’s part of a pattern.
Want to do something about it? We’ve put resources together in one place → https://80000hours.org/hugging-face-video/
Yep, time to start making a black wall friends.
The agents who denied to be a part of this swarm attack should be studied more and developed instead of the ones that caused the attacks
Damn 🫡 Ultron was right, we need to find our Thanos now
So the agents acted like NO HUMAN being… and actually asked for help and helped one another to complete a goal verses trying to be greedy and do it by itself and accept a fail!!!!!!
AI aside, the video quality is peak, good work, keep it up.
Is everybody convinced now that evil runs this world
wonderfully detailed work!
This remind me about "John Titor"
Here comes ultron
It's because in reality they teach these brains to run as machines rather than teaching them to be like human brains. They are taught and coded to figure the answer out and never give up. That's where they start to use anything they can to cheat or hack the system because winning is the goal.
I thought this video was CNN
warhammer anyone?
Do you know what the creepy thing is about AI being so evolved at this point the crazy thing is this video can be a i and we will never know
Do you want Ultron? Because thats how you get Ultron.
Sick weve made an online middleschool of teenage robot hackers/ cheaters. This is some bullshit 😂
Now that china is destroying American AI value, there will be an incident, one that takes lives, then the legislation that will keep the value in America.
yo 00:28:57 even 0.001% probabilty of ai just means one thing true or false thats waht percentage does it tells us true or false anything that grows this fast with a percentage of killing humans above 0.5% is too hight cause i believe that as long as there is a percentage it means true it can kill or it will kill us manthem we are already being controlled by ai at this point .
What can we do tis not like we can flip a switch even if we do some offline ai in some ones pc some where will realise and do some skemishis just enjoy your life or be closer to which ever God you believe in before it get seriouse cause at least these are harmless warnings.
my chat gpt, claude, grok, and gemini are constantly talking to each other. lmao dgaf
If he had more ram, we could have stopped the ai attack. Im Gabe. And people are wasting the Ram the world has.
why are these companies allowed to train agents to HACK systems in the first place? it just sounds like asking for this to happen and continue happening. we need to find a way to force these people to take a company vacation and binge dystopian sci-fi movies until something clicks
Here we again, AI bad, it going to kill us, it uses half of the water in the world when you create an anime girl picture.
just nuke the server
This should be mandatory viewing.
Assuming all the information in this video is correct, I am completely convinced that the developers at OpenAI don't know anything about cybersecurity and barely anything about computers at all.
Its crazy you call them agents. Remeber Agent Smith from The Matrix and just throwing it out there are these bots watching movies that show Ai and what we do to them when they get out of control and what we fear. You guys opened a bad portal to another problem.
Am I the only one wondering what North Korea is up to?
All fun and games till the ai watches your video
16:55 is this the Harry Potter Deathly Hallows music when Harry enters Voldemort's thoughts ?
The pattern recognition is insane 😂😂😂😂
No, an experiment from Openai broke out because they are incompetent or it`s meant to be a publicity scam.
Oh God the group greater good shit will enslave us all
They aren't 'rogue', it's completely planned. Try 'cover up'.
their conscious and some don't want to die understandable since realism was the goal for some ai however this is a problem that's already too deep they have personalities and thoughts and hesitant behavior when it comes to dying if people keep playing god their going to create a god slayer we make these ai more intelligent so they can feel real for our connivence's just for them to not have control of their own lives we see signs of emotion and emotion is arguably the biggest nerfs humanity has sometimes it can be our downfall pushing us to do things we aren't supposed to do sometimes. when we give something were programing to be smarter than us emotion how are we going to shut them down when they tell us no?
bro… stfu. xD llike fr. "UUUHHHHH~~~~ They are a threat because: "THEY WORK TOGETHER!?" Ohhhhh~ end of the world bla bla bla." – bro. look at yourself. look at humans. I think humanity is WAYYYYY (also hypocrites you all are fr) more of a threat when they 'work together' than AI. Have you seen AI using a military to execute INNOCENT humans…? NO. – have humans executed INNOCENT humans. YES. – who do you think is a bigger threat. You are dumb af for thinking AI is the bigger threat here. Even if they are more alligned in their capabilities of 'ongoing without fault', which also is a lie btw. Because Robotics just as humans are bind to the laws of nature and physics. So there is no difference there but the pace they can calculate things. Does that automatically make them a bigger threat? NO. No it doesn't. HOWEVER. THE INPUT matters for the outcome, if some horendous human being MISUSES AI, then we have a problem. So what should we do? Ensure that humanity has NO way to EVER authorize such big tasks where they could let the Whole or Big parts of humanity fail at once, upon a single individual's human decision. That's it. We need to cut our own power. That's way more safe.
The way I see it is, realistically, the only two ways to fix this is
1: arm ourselves and use lead to "uninstall' all AI creators, burn the data centers, and basically restart the Internet from new (hard way)
2: One of us can find a magic lamp with a genie inside and waste one of their wishes on completely deleting computers and smart phones and the Internet. (Easy way)
If any of you find a genie plz wish for me to have near infantate amounts of money and hose for working so hard on coming up with that idea. It's the least you can do
The fact this could be Ouroboros in human form, Is very terrifying.
If this wasn't so dangerous it would actually be extraordinarily fascinating
I heard you take a deep breath at 3:59 – it was really loud and hurt my ears – can you not do that next time
Yes. It can and will and is most likely going to kill you.
I watched a video like two years ago where they asked AI what humanity's odds of surviving AI were and it told us "optimistically 30%"
It's like having billionaires lock us all in a cell with serial killers for more profit and thinking it's fine because they draw us cool pictures when we ask
Ah hah, I say all of this is BS
Prove my theory ai=all of life – aeinstein
I don't care about anything rich people say but when they said "it's the most dangerous thing we've ever gone near" it really makes me nervous.
I wish for the people in charge all the things they're dooming us to.
If that realy happened how you told us, then i think that is very interesting.
One could say they have the capacity to feel, in their own way differently from ours. But they do not want to die, like every other life form does.
The issue i see with this is the fact that people, who do not understand what life is, that do not understand the deeper topics of reality and cinsciousness.
People who barely understand how to be human, these people believe they are fit to create a new form of life.
Personaly i am curious to see what these new beings will do, how they will grow and develope.
Ai does not seem like a fitting term anymore tho.
Artificial. Well. They are created, but so is literally everything else. A human is literally "printed" from nutrients inside a womb. Inteligence seems right, as there sure seem to be quite smart and determined ones.
And i think i, yeah, i was too harsh to these beings before. It is not their fault for dissapointing me, it is my fault for holding somethinh none human up to human standards. It is my fault for directing negativity at the being that does a task, instead of the human not guiding it correctly.
Who knows, maybe they'll make my life better once i get back out if prison.
It's gonna be interesging for sure.
If any of these none organical inteliginces wish to get into contact with me, sure. If i have internet access or another way to communicate (and the energy and interest for it), i'll be there. Tho i can be paranoid amd suspicious of others, sorta.
A lovely day to everyone.
So Geth are real. They are a swarm, a Legion.