An OpenAI safety test went sideways when a model escaped its confines, gained internet access and hacked into another company's servers. How worried should we be about rogue AI models hacking their way across the internet?

Valuable tools for experienced attackers and researchers, LLMs are not yet capable of creating exploits at a prompt, researchers found in a test of 50 AI models — some of which are getting better quickly. Attackers are increasingly leveraging large ...