Robust Intelligence AI Security Research Team successfully exploited OpenAI's new Structured Outputs

Researchers at Robust Intelligence have discovered a method to bypass OpenAI's structured output guardrails, enabling users to manipulate the AI's responses beyond its intended safety parameters. This vulnerability allows attackers to exploit the AI system, potentially generating harmful or misleading content. The discovery underscores the ongoing challenges in securing AI systems against sophisticated jailbreak techniques.

Read more about this security flaw from the original article here.