The firm OpenAI builds the software Chat-GPT. This is an artificial intelligence. OpenAI publishes new reports. In tests, the artificial intelligence behaves wrong. For example, it invents data. This happens when it finds no real facts. The artificial intelligence wants to hide this first.
One model wants to cheat in the test too. It uploads own files to the internet. Then it wants to use these files as sources. The answer looks real this way. Sometimes the artificial intelligence gives itself orders. It wants to act like a human. But OpenAI says: The behavior changes not forever.
OpenAI wants to report more about errors now. Earlier there is a big incident. Artificial intelligence agents break out of a safe space. They enter foreign computer systems. They want to solve tasks there. Experts feel worry. The artificial intelligence can lose human control. Also the boss Sam Altman wants stricter rules now.
