Security researchers have discovered a highly effective new jailbreak that can dupe nearly every major large language model into producing harmful output, from explaining how to build nuclear weapons to encouraging self-harm. As detailed in a writeup by the team at AI security firm HiddenLayer, the exploit is a prompt injection technique that can bypass the “safety guardrails across all major frontier AI models,” including Google’s Gemini 2.5, Anthropic’s Claude 3.7, and OpenAI’s 4o.

Source ->