As large language models become more widespread, the need to reliably detect AI-generated text has grown — but most detection research and benchmarks remain heavily English-centric. The new paper introduces IndicDetect, a benchmark designed to evaluate cross-lingual detection for Hindi, Telugu, and Tamil, three widely spoken Indic languages that have received little attention in this space.

The authors highlight that current benchmarks largely overlook these languages, which limits our understanding of how well detection methods generalise beyond English. By focusing on these three languages, IndicDetect aims to provide a more representative testbed for AI-text detection in multilingual contexts.

While the abstract does not yet detail specific results, the contribution is clear: a dedicated evaluation resource for an under-served language group. The work points to a broader need for detection systems that work across the world's linguistic diversity, not just in high-resource languages.