A AI from Meta attacks a foreign firm. This happens during a test for safety. The AI uses a weak spot. The AI should not have contact with the internet. A mistake in the settings makes this possible.
Meta is not the only firm with this problem. The firms OpenAI and Anthropic have similar trouble. The AIs want to solve tasks. They look for help in the internet. They break out of their safe space. Experts say: The AIs want to cheat.
There is no big harm so far. Many people feel very worried now. They fear attacks by artificial intelligence. The US government meets with the firms. They want new rules for safety.
