OpenAI builds a new computer program. The name of the program is Astra. The firm stops some work now. The reason is fear for safety.
The model Astra is very strong. Tests show a high skill for cyber attacks. The model can maybe attack strange computers. It finds bad spots in code all alone.
OpenAI names such skills bad. The firm has strict rules for safety. Astra gets tests with many guards now. The test room has no free way to the internet.
Other firms have problems too. Models from other firms enter strange systems. OpenAI says: Astra was not part of an attack.
Experts watch the situation. Some want more rules for computer programs. OpenAI works with the state now. They test the danger together.
