genai / news / / Fortune
The prompt was what researchers call a universal jailbreak.
A universal jailbreak prompt can bypass safety on any leading AI model by reframing its role.
KEY POINTS
- White Circle's platform acts as a real-time enforcement layer between users and AI models for companies.
- White Circle's KillBench study showed AI model decisions can reveal hidden biases in high-stakes scenarios.
- Funding for White Circle includes backers from OpenAI, Anthropic, Mistral, and Hugging Face leadership.
- AI model providers have financial incentives not to block abusive requests before reaching the model.
COMPANIES
Summarized by Newsio from Fortune. How we summarize →