Researchers · Interviewed at NeurIPS 2025

Sara Price

Anthropic

NeurIPS 2025 in San Diego. We Jailbroke Every Frontier Model

NeurIPS 2025 · December 19, 2025 · 10 min

NeurIPS 2025 in San Diego. We Jailbroke Every Frontier Model

How robust are today’s frontier AI models—really?

In this post #NeurIPS2025 episode, Sara Price (AI safety researcher at @anthropic-ai ) walks us through a surprisingly simple black-box attack that successfully jailbreaks frontier text, vision, and audio models—often with very high attack success rates.

More from NeurIPS 2025

All event interviews →