This study extends mindset theory to language learning with AI by developing and conducting a preliminary validation of the AI Language Learning Mindset Inventory (AILLMI). Building on Dweck's concept of growth and fixed mindsets, and incorporating Lou et al.'s (2017) decremental mindset, we explore how these constructs manifest in learners' engagement with AI language learning tools. The AILLMI was developed through a rigorous two-stage process. An initial 18-item pre-pilot (n=14) informed the design of an extended 79-item instrument. The main pilot study (n=66) provided data for comprehensive Rasch analysis, which was central to our validation process. This analysis led to the refinement of the AILLMI, resulting in a revised survey instrument with strong psychometric properties. Our findings indicate that AI language learning mindsets extend beyond the traditional growth-fixed dichotomy. The inclusion of decremental mindset items provided new insights into learners' concerns about potential skill deterioration with AI use. The Rasch analysis also revealed patterns in item difficulty, with items related to personalised AI feedback and expanding AI tool use being the most challenging for learners to endorse. We will present the current version of the AILLMI, discuss its structure, and explore its potential applications in research and educational settings. While further refinement of the AILLMI is anticipated, this instrument represents a significant advance in our understanding of, and ability to measure and shape, AI language learning mindsets, with implications for the development of AI-integrated language learning environments.