Over 2,800 participants from around the world contributed more than 600,000 adversarial prompts aimed at manipulating three state-of-the-art LLMs.