Frontier models refuse to help organizers, so we built our own activist AI (www.outcryai.com)

Outcry has unveiled a new activist-focused artificial intelligence model, designed to operate locally on older smartphones, emphasizing privacy and usability for grassroots organizers. This model integrates four distinct conditioning layers—quantization for compactness, a LoRA adapter for domain-specific knowledge, Contrastive Activation Addition (CAA) for real-time persona steering, and a soft prompt to enhance model stability. Each layer can be independently updated, enabling greater flexibility without requiring retraining of the base model. This initiative is significant for the AI/ML community as it pushes the boundaries of how small-scale AI can be tailored for specific domains, particularly in activism and organizing. The model is purposely designed not as a general AI but as a focused tool for activists, addressing the unique challenges of movement work while prioritizing data privacy—ensuring that conversations remain on user devices without server-side involvement. The collaborative potential of Outcry is clear; it invites research partnerships to refine its technology and offers support for organizations looking to build customized AI solutions based on their proprietary data.