As Generative AI (GenAI) becomes increasingly embedded in language classrooms, many EFL learners risk becoming passive consumers of algorithmic outputs, often deferring to AI’s perceived “native-like” authority. Such reliance can undermine learner agency and reinforce standard language ideologies. This study reports on a classroom-based pedagogical intervention, Speaking Back to AI, designed to help Japanese university EFL students critically engage with GenAI tools. Drawing on the APSE model of Critical AI Literacy, the intervention was implemented as a short module within a regular EFL course. About 50 Students participated in iterative “red-teaming” activities, evaluating GenAI-generated content for hallucinations, linguistic stereotyping, and pragmatic misalignment with local norms. Rather than revising their work to align with AI suggestions, students produced multimodal artifacts, such as annotated AI outputs and short reflective digital products, that corrected, resisted, or recontextualized problematic outputs. Data sources include screen-recorded red-teaming tasks capturing students’ testing and negotiation with GenAI outputs, reflective journals, and pre- and post-intervention surveys. Preliminary observations suggest increased learner awareness of AI limitations and greater willingness to assert authorial voice when negotiating with GenAI outputs. The presentation concludes with reflections on challenges and how similar Critical AI Literacy interventions can be adapted for other EFL contexts.