Language technologies are increasingly ubiquitous and now translate emergency bul-letins, draft clinical notes and mediate everyday conversations, yet their impressivefluency can be misleading–masking limited reliability, unpredictable errors and unevenperformance across different user groups and languages. Building on Shneiderman’shuman-centered AI (HCAI) paradigm, this article introduces the Human-Centered AILanguage-Technology (HCAILT) model, a domain-specific framework that binds reli-ability, safety culture and trustworthiness to the full language-technology pipeline.HCAILT couples technical guardrails (such as retrieval-augmented generation and qual-ity estimation) with organizational practices (like bias audits and incident-reportloops), together with user-facing features that maintain meaningful human control.Two blueprint use cases–in multilingual healthcare and crisis communication–illustratehow the HCAILT model guides system architecture, deployment practices and evalu-ation. A demo system demonstrates immediate feasibility on public large languagemodels. By translating HCAI principles into actionable design levers, HCAILT providesscholars, developers and policymakers with a pragmatic path from ethical aspirationto deployable practice. The paper concludes with a research agenda for empirical val-idation in real-world settings and invites multidisciplinary collaboration to ensure thatnext-generation language technologies are not merely powerful, but demonstrablyreliable, safe and worthy of public trust.