An AI-generated answer can reproduce an official figure correctly and still misrepresent what it means. A comparison may overlook a change in definition, a summary may omit uncertainty, or an explanation may imply a causal relationship that the data do not establish. Such examples raise questions about how statistical quality is preserved when AI tools become part of the interpretation and use of official statistics. Building on the relationship between statistical literacy and quality, as explored at ISI 2025, this presentation will examine the role of human judgement, critical thinking, and ethical responsibility in AI-assisted statistical work. It asks how literacy can help producers, intermediaries and users to recognise the limits of automated outputs and make informed decisions about their use. The discussion is organised around three responsibilities: checking the evidence, assessing the interpretation and accounting for the resulting claims. These involve verifying sources, definitions, reference periods and revisions; examining comparability, uncertainty and the strength of an explanation; and being transparent about the contribution and limitations of AI. They also require attention to whose experiences are represented, which groups may be obscured by aggregate results, and whether a technically correct statement could lead to a misleading or unfair interpretation. Ethical responsibility extends beyond identifying factual errors. Who should verify an AI-assisted interpretation before it is communicated or used in a decision? What should users be told about how an answer was produced? How can statistical organisations encourage innovation while ensuring that responsibility remains identifiable? These questions connect statistical literacy with transparency, explainability, fairness and accountability. Illustrative scenarios involving official statistics will provide a starting point for reflection. Drawing on concerns relevant to statistical communication and education, the presentation considers how explanatory metadata, staff development and learning activities, including capacity building and statistical literacy, can help people question and justify AI-assisted answers. It also asks what support statistical offices should provide, recognising that users differ in their knowledge, resources and ability to check automated claims. Rather than treating statistical literacy as an individual skill alone, the presentation considers it a shared responsibility involving statistical institutions, educators, technology providers and users. Its purpose is to open discussion on the competencies and institutional safeguards needed to benefit from AI while preserving statistical understanding, ethical judgement, critical thinking, and good practice. A central question is how to distinguish tasks that can reasonably be delegated to AI from judgements for which people and institutions must remain accountable.
