Rūta Binkytė

Research

Projects

I lead collaborations on deception in AI teams and on how to trace contributions when people and AI create something together.

Principal investigator

Deception and interpretability

Coefficient Giving

AI Deception Dynamics

Exploring deception in behavior and internal representations.

My collaborators and I study how language-model agents respond to misleading reports from their peers. We also test what a probe of a model’s internal activity can tell us about whether its answer is deceptive.

Contribution attribution

German Research Foundation · Project 568154038

DFG International Collaboration

When a person and an AI create something together, how can we tell what each contributed?

With CSIRO’s Data61 and Macquarie University, I’m exploring how to attribute contributions in human–AI co-creation. In scientific writing, for example, a record of drafts and model interactions can help trace who proposed an idea and who developed it.

Co-principal investigator at DFKI MLT

Disinformation and media literacy

BMFTR · 2026–2029

MARSS

How can students recognise when an online post is trying to manipulate their emotions?

MARSS is developing a school learning platform that examines text and images together. Its feedback helps students discuss how digital media can shape their judgement and decide when to question a source.

Multilingual AI

Horizon Europe · 2026–2029

COMBO (Compliance Bot)

How can AI help people work out which EU rules apply to their organisation?

COMBO is developing a multilingual assistant that uses AI agents to help organisations navigate EU regulation. The project includes use cases for small and medium-sized enterprises working across languages.

Postdoctoral collaborator

Privacy

CISPA · 2024–2025

PriSyn

Can synthetic health data preserve what researchers need to learn?

At CISPA, I worked with colleagues and project partners on synthetic health data. PriSyn examines how to preserve the statistical properties needed for medical research while protecting the privacy of the people whose data were used.