Jailbreaking AI

Jailbreaking AI

'Jailbreaking' AI models to teach you how to make Molotov cocktails or worse is a dangerously simple process.

The method known as “Skeleton Key” poses an increasing risk to AI models since it takes advantage of holes in existing security measures to retrieve private data. By avoiding security features in models like Meta’s Llama3 and OpenAI’s GPT 3.5, this technique—which was brought to light