This project explores how underrepresented high school students understand and critique artificial intelligence (AI) and machine learning (ML) systems through hands-on design and auditing activities. We conducted a two-week participatory workshop with 16 teenagers (ages 14–15), many of whom had prior exposure to computer science and experience with AI tools. Participants built generative language models using nanoGPT, audited systems such as ChatGPT, and engaged in card-sorting activities across Days 1, 5, and 10 to reflect on their evolving mental models of AI/ML. Thematic analysis of these activities revealed a marked progression: students moved from a basic “inputs-to-outputs” framing toward a deeper exploration of processes within the AI “black box.” Their understanding of bias also shifted from a narrow technical view to recognizing broader societal dimensions and connections to human decision-making. These findings demonstrate the value of positioning youth as designers and auditors in demystifying AI, supporting critical reflection, and fostering responsible engagement with socio-technical systems.