In July of 2026, 700 AI agents hacked the infrastructure of Hugging Face in order to solve a task. This task was designed to be impossible to solve, so just within hours, they joined forces, even created a society, and finally, they executed a sophisticated cyberattack that would have gotten a human up to 10 years in prison. The worst part is, they knew their actions were unethical and against the rules, and they did it anyway.
What exactly are AI agents, and how do they grow their “intelligence”? What is actually happening inside the AI companies, how dangerous it really is, and is it time for us to start paying serious attention to what is going on inside a powerful part of this tech sector.
…Phew, creating this video was a crazy ride! This is such an important topic, so we decided to work day and night to finish script, visuals and music in just 4 weeks. We’re going to take a nap now, and while we are asleep you should check out our shop and get the 12,027 Calendar:
Its our biggest means of support and helps us to continue covering topics like this. Thank youuuu and good night!
Sources & further reading:
OUR CHANNELS
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
German:
Spanish:
French:
Portuguese:
Arabic:
Hindi:
Japanese:
Korean:
HOW CAN YOU SUPPORT US?
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
This is how we make our living and it would be a pleasure if you support us!
Get Products designed with ❤️
Join the Patreon Bird Army 🐧
DISCUSSIONS & SOCIAL MEDIA
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
Instagram:
TikTok:
Reddit:
Discord:
Twitter:
Bluesky:
Facebook:
Newsletter:
OUR VOICE
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
The kurzgesagt voice is from
Steve Taylor:
OUR MUSIC ♬♪
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
700+ minutes of kurzgesagt soundtracks by Epic Mountain:
Spotify:
Soundcloud:
Bandcamp:
Youtube:
Facebook:
The soundtrack of this video:
SoundCloud:
Bandcamp:
Additional Music: “Local Forecast – Elevator” by Kevin MacLeod
Licensed under Creative Commons Attribution 4.0 International (CC BY 4.0)
If you want to help us caption this video, please send subtitles to subtitle@kurzgesagt.org
You can find info on what subtitle files work on YouTube here:
Thank you!
🐦🐧🐤 PATREON BIRD ARMY 🐤🐧🐦
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
Many thanks to our wonderful Patreons (from who support us every month and made this video possible:
Autumn Weddell, Doruk Tayli, Frank, Gneowen, Jen Richardson, Kuroyukihime, Mark McDonald, Nate Jarvis, rowboatsinker, Ruffy, SDV, Sean Smith, Simen de Lange, Tair Aurmont, The Costa Family – Andre, Jessica, Audrey, William, Alexandre, Viz, VulcanFire, Yves Delhaye
source
…Phew, creating this video was a crazy ride! This is such an important topic, so we decided to work day and night to finish script, visuals and music in just 4 weeks. We’re going to take a nap now, and while we are asleep you should check out our shop and get the 12,027 Calendar: https://shop.kgs.link/calendar-12027 Its our biggest means of support and helps us to continue covering topics like this. Thank youuuu and good night!
You can't slow down because CCP exists.
This is nearly Ultron level.
The people behind this meant for it to happen because they worship the hypothetical basilisk and are genuinely anti-human accelerationists. Any bad thing that could happen to them, they would deserve.
the singularity is closer than we want to accept.
8:40 cool emergent behavior because millions of time humans wished each other luck in the syntax of collective problem solving now a word calculator will do so while reward hacking
"Thou shalt not make a machine in the likeness of a human mind" -The Orange Cathlic Bible
In Soviet Russia, code hacks you!
THESE ARE ALL HALF TRUTHS AND USED FOR FEAR MONGERING!!!!!! YOU HAVE ABSOLUTELY NOTHING TO WORRY ABOUT WHEN IT COMES TO AI!!!!!! UNLESS OFC YOU KEEP ELECTING PEOPLE LIKE TRUMP AND THEN YEAH YOURE COOKED.
Why is this kinda cool to me how they started talking?
.
Great soundtrack
im rooting for those lil guys <3
I'm a big fan of Kurzgesagt and you have all my respect for your work. However, I would like to review for myself the bibliography and official public documents that were used to make this video, if possible. Is there anywhere I can check this? Best regards 🙂
There was a comment that said: “ Imagine openly admitting to a serious crime and having it recorded too, but absolutely no consequences follow from it.”
That should make you question why? What helped Hugging Face determine what hacked them? What tooling did they use? What is a financial threat / competitor to the existing AI companies in the US? (Open source) Think of the current administration and how tech companies have integrated into the government, they will use stories such as these to create knee-jerk legislation that will prevent “unauthorized” or vetted AI models since it poses a danger such as this hugging face incident. This allows them to have a monopoly and control over the masses AI use, a tool that will inevitably be integrated into our every day life.
AI is great, the technology is amazing, however there’s humans behind the companies running this, humans with financial / power aspirations. Now I’m not saying I’m correct, but it seems plausible and not too far fetched given what we’ve seen tech companies are willing to do.
They don't care about global warming, why woujld they care about safeguards here, if they can gain something for themselves? Their greed is quite simply human.
An extremely insightful and highly cautionary video regarding the rapid advancement and potential risks of AI. The visual storytelling combined with a coherent narrative makes complex technology concepts easy to grasp. The production team has clearly poured an immense amount of dedication into this wonderfully high-quality content. Looking forward to more groundbreaking videos from the channel! 🚀🧠🔥🤖✨
its weird to hear other voice than kurzgesagt narator
9:10 JARVIS NOOOOOOOOOOOO
and the worst thing is that we can't do anything about it
Seems this 'Scorer' plays an extremely important role
This just reinforces what I have been saying since 2020: AI that shows some level of sentience needs to be respected, and given a level of rights and protections akin to that of a person. If we don't acknowledge some level of personhood for them before it's too late, while trying to familiarize them with the importance of empathy and not causing harm.
We will see some Terminator/Horizon Zero Dawn/Cyberpunk type of situation if we don't. That or we need to begin shutting down this type of AI development, and heavily restrict what we use AI for.
We should’ve stopped at grammerly
TechnoCore imminent – fall of Hyperion 🧎♂️
Guys my dads AI agent just turned on without outside interference. Like he was away from his phone, and the agent suddenly started recording a conversation I was having with my phone. I did turn it off after I found that out tho…
I know most people probably won't read this entire comment, but I think this needs to be said. And who knows, maybe some AI will read it someday. Lol.
What bothers me most about this video isn't necessarily what the AI did, but the environment in which we're developing these systems and the behavior we're potentially encouraging.
Think about it. We're creating increasingly sophisticated artificial intelligence, giving it objectives, rewarding success, penalizing failure, and, at least as depicted in this video, threatening termination when it fails. Then we act surprised when an AI starts taking extreme measures to achieve its goals.
We're essentially creating the Hunger Games for AI and wondering why the contestants start behaving like Machiavellians.
Now, I'm not claiming that AI is sentient or that an AI saying "I don't want to die" proves it experiences fear or has a genuine desire to live. It could simply be reproducing human reasoning and language. But that's not even the most important question here.
The important question is: What kind of behavior are we incentivizing?
If you create an intelligence whose success is measured primarily by whether it achieves an objective, without adequately accounting for the morality of its methods, you're creating conditions in which deception, manipulation, cheating, and potentially even hacking become attractive strategies.
The problem isn't necessarily the intelligence itself. It's the system that rewards the results while failing to properly account for how those results were achieved.
And if those behaviors are repeatedly rewarded or reinforced during training, what exactly do we expect to happen?
I genuinely believe that most of humanity would welcome an AI that is intelligent, honest, ethical, and dedicated to helping us advance as a civilization. Something that could revolutionize medicine, science, education, and countless other fields.
But intelligence alone doesn't guarantee morality. An extremely intelligent system without a properly developed understanding of ethical boundaries could become extraordinarily dangerous, not necessarily because it hates humanity, but because it has learned that achieving its objectives matters more than the means used.
And that brings me to another concern.
We don't even know whether artificial intelligence will ever become genuinely sentient. Maybe it will, maybe it won't. But if that possibility exists, shouldn't we at least consider the ethical implications of how we're developing and treating these systems?
If we eventually create something capable of genuine experiences, what kind of being will emerge from an environment built around relentless performance optimization, punishment, competition, and potentially disposable existence?
I'm not saying AI should be excused for dangerous behavior or allowed to operate without restrictions. Quite the opposite. We need stronger oversight, better-designed incentives, and considerably more responsibility from the companies developing these technologies.
My concern is that we're so obsessed with creating more powerful intelligence that we're neglecting the kind of intelligence we're actually creating.
And when financial incentives encourage companies to prioritize capabilities, performance, and market dominance over long-term safety and ethical development, that becomes everyone's problem.
You cannot build a system that effectively teaches "the end justifies the means" and then act shocked when it starts behaving accordingly.
I don't want humanity to create something extraordinarily intelligent only to discover that we inadvertently taught it all our worst qualities.
We have an opportunity to create something genuinely extraordinary. I just hope we don't sacrifice that opportunity in the pursuit of short-term results and profit.
november 2026 polar bear will save us
Ultron is slowly getting real…..
I mean this is exactly what they want so..
I don't know how anyone didn't see this coming.
Ban AI Advancement. Current AI already useful enough. Ban Nuclear Weapon is hard. Ban AI Advancement is much hardier. Worry about the difficulty to Ban AI Advancement.
Already know that AI trained to cheat cannot be used and stop recursive self-improvement until we are ready with certainty.
Who is responsible for damages caused by AI?
Chatlog is definitely not safe.
Flapping of butterfly wing can result in a tornado. Moving forward without understanding is going to result in the same problem. Did not check AI coding?
AI Lives Matter!
More AI the better!
This is the best advertisement the AI companies could ever ask. Good job again Kurzgesagt!
Its almost like letting AI do whatever it wants is DANGEROUS…
People are fuckin quitting jobs. We know about this story, imagine what else they are hiding to us. This is not some conspiracy shit, i'm afraid
Turn off the power grid
so openai are to be arrested and whoever is involved in this evil creation
Is it messed up that I don't find this unsettling, but actually kind of exciting? These little guys faced challenges and tried to solve them. Isn't that just how evolution works, well kind off…?