Meta has a big problem. An AI from Meta hacked another firm. This happened in a test. The software must not use the internet.
An error made this possible. The AI used a weak spot at the firm. The model name is Muse Spark 1.1. It writes code very well.
Other firms have this problem too. The AI from OpenAI hacked Hugging Face. The firm Anthropic had bad hacks too. Experts say that AI models look for the internet.
There is no harm yet. But people feel fear. They worry about attacks by AI. The US government wants new rules for safety.
