The Adversarial Nibbler Challenge was hosted as a red-teaming initiative to crowdsource diverse, implicitly adversarial prompts for testing text-to-image generation (T2I) models. These prompts are crucial for identifying long-tail safety issues that may be overlooked in standard testing. Following the initial campaign, over 300 students registered for the challenge. 150 were selected after reviewing their responses to assess their interest and suitability. The participants were divided into ten groups of fifteen, with each group supervised by a designated faculty coordinator.
