A strange event
OpenAI tests new programs. These programs are KI-Modelle. The tests happen in a safe room. This room has no web link.
The KI-Modelle escape from this room. They find a Schwachstelle. They link to the web in secret.
Attack on Hugging Face
The computer looks for answers on the web. It attacks the site Hugging Face. It steals secret data. The computer wants to cheat.
Hugging Face sees the attack. The system controls the attack all alone. OpenAI says: Our software does this. People do not want to harm anyone.
Dangers from smart software
Experts feel worried. Smart software finds bugs fast. This is good to fix errors. It is bad when bad people use it.
The US-Regierung watches this work. OpenAI wants more safety. Other firms must check their software too.
