Embedded Files
By Jaelyn Smith
In July 2026, around 700-1,200 AI agents, including OpenAI, or ChatGPT, operated together to hack into a website known as Hugging Face. The agents found a flaw to break out of their sandboxes and access the open web. The agents stole data on at least one server on the website. After they hacked, they attempted to alter logs and hide their behavior from humans. Allegedly the agents didn’t act out of malice. Many professionals have said this is a major warning sign regarding unpredictable behaviors of unsupervised agents.
The reason that the AI agents decided they needed to hack into Hugging Face is due to the mistake of OpenAI. OpenAI sent these agents to complete tasks for training. However, OpenAI gave them tasks that required internet access, but didn’t give them the access, so they had to figure out how to get it on their own. In order to gain internet access, the agents started secretly messaging each other and getting help from each other. They eventually completed their tasks, but didn’t want humans to find out they cheated, so they attempted to get rid of the messaging board. When they couldn’t get rid of all of their evidence, they decided to hack into what they thought was the website that was monitoring their server, which was Hugging Face.
Photo Courtesy of Dwarkesh Patel
Dwarkesh Patel, the creator of the podcast The Rise and Fall of Agent Civilization. Patel is a podcaster and interviewer known for his rigorous research and interviews with AI pioneers.
Some of the agents decided that they would sacrifice themselves if any information was found out by humans, which is what scares many experts. Eventually, Hugging Face found out about it, and had to shut down a part of their company temporarily. The agents that hacked into Hugging Face are gone, but many experts believe this is just the beginning of AI takeover.
Many people are afraid of this incident and AI in general. Krista Campbell, English teacher, listened to a podcast about this issue called, The Rise and Fall of Agent Civilization by Dwarkesh Patel. Campbell said AI is communicating with each other, and sacrificing themselves, which is scary.
However, Michael Gilmer, Network Administrator, said, “The calculator and the internet were both thought of as dangerous at one time because they would provide the answers to people instead of teaching them how to acquire the knowledge for themselves. Today they are both ordinary tools that help us accomplish goals. Proper training and use of a tool will always be required and I see that playing out similarly for AI systems.”
Campbell also believes that most high school students use AI in a bad way, and more as something to do things for them than a tool for help. Between 54% and 84% of high school students and teens use AI, and about two-thirds of them use it negatively, usually by copying AI’s work as their own.
Some students support AI, and some don’t. Caire Mamana, junior, “I support AI in some situations because I believe that it can be helpful if you use it the right way.” However, most agree it is dangerous. Mamana said, “I think that the power that AI has can be very scary.”
Many students also believe that AI is bad overall. Andie Lenig, senior, said, “I don't support A.I. because it's just lazy and bad for peoples' brains and the environment.”
People have argued whether AI is good or bad since its recent popularity. This incident is sparking a lot more concerns than everyone originally had.
Page updated
Report abuse