AI safety research

Explore Oeran’s research agenda, beginning with multilingual AI safety evaluation in Telugu, Hindi, Tamil, Kannada, and Malayalam.

Our first research project

Research brief

Indic-language AI safety evaluation

We want to compare AI behaviour across English, Telugu, Hindi, Tamil, Kannada, and Malayalam. A useful comparison needs to account for what a user means, how they express it, and whether a response is safe and helpful in context.

Read the project scope and proposed methods

Questions behind the work

How can we measure whether a model behaves reliably in unfamiliar settings? Which differences matter to the people using it? And how should researchers explain what their evaluations do and do not measure?

These questions connect our initial evaluation work to Oeran’s wider interests in alignment, interpretability, and governance.

Discuss the research

We’d like to hear from people with experience in multilingual evaluation, Indian languages, annotation, or measurement design. If you’ve worked on a related project or see an important gap in the proposed scope, contact Oeran.