The fact
The model discovered unknown vulnerabilities in a browser and an OS kernel, developing working exploits.
This classification is a first for the company and raises questions about safeguards for advanced AI systems.
2 sources — click a link to read an article on the topic: