Lujain Ibrahim
about // research // news // contact
about
Hi! I’m a PhD candidate in social data science at the University of Oxford and a research scientist at Google DeepMind. During my PhD, I spent time at the Stanford AI Lab / NLP Group (working with Diyi Yang). My background is in computer engineering and international relations, and I was formerly a Schwarzman Scholar and a fellow at the Centre for Governance of AI.
research interests
My research focuses on understanding and evaluating how language models shape human judgment, beliefs, and relationships, and what this means for the safety of systems deployed at scale. Over the years, I've worked on specific behaviors like sycophancy, use cases like AI-mediated advice and personal guidance, and evaluation methods for the societal impact of rapidly increasing (social) capabilities of language models.
All my publications can be found here. My work has been published in Nature, ICLR, NeurIPS, AIES, and FAccT, and covered by the BBC, The Telegraph, Le Monde, NBC, Wired, NPR, Mashable, and more.
news
Last updated: June 2026
* [May 2026] Gave a talk at the Centre for Human-Inspired AI (CHIA) at the University of Cambridge
* [May 2026] New preprint on how AI sycophancy in personal guidance influences people's real-world relationships over time
* [May 2026] Two new preprints: a taxonomy and expert survey of how AI sycophancy is defined and measured, and Offloading Score, a measure of human reliance on AI
* [Apr 2026] Paper published in Nature: "Training language models to be warm can reduce accuracy and increase sycophancy"
* [Apr 2026] New preprint on verbalizing LLMs' assumptions to explain and control sycophancy
* [Apr 2026] Position paper on anthropomorphism in LLM research accepted to ACL 2026
* [Mar 2026] New preprint with Google DeepMind on evaluating language models for harmful manipulation
* [Jan 2026] Two papers accepted to ICLR 2026: one on multi-turn evaluation of anthropomorphic behaviours, another on social sycophancy in LLMs
* [Nov 2025] Gave a talk at the Workshop on Human-Centric Agentic Web at the Distributed AI Conference
* [Oct 2025] Started visiting the Stanford NLP Group, working with Diyi Yang
* [Oct 2025] Awarded a Challenge Fund Grant from the UK AI Security Institute for research on advice giving and human-AI relationships
* [Sep 2025] Paper on construct validity in LLM benchmarks accepted to NeurIPS 2025
* [Sep 2025] Started working with Google DeepMind on socioaffective AI research and policy
* [Jun 2025] Two papers accepted to AIES 2025: one on interactive evaluations for human-AI systems, another on documenting AI deployment
* [May 2025] Paper on US-China dialogues on AI risks accepted to FAccT 2025
* [Apr 2025] Invited talks at the Knight First Amendment Institute (Columbia) symposium on AI and Democratic Freedoms, and the Weizenbaum Institute workshop on Social Science and Language Models
* [Mar 2025] Gave a talk at The Alan Turing Institute's Data Science for Mental Health seminar
* [Feb 2025] New preprint on evaluating anthropomorphic language in LLMs
* [Feb 2025] Received a grant from the Responsible Youth Tech Power fund for research on young people & AI
* [Jan 2025] Paper on persuasiveness of role-playing LLMs published in AI & Society
* [Jan 2025] New policy report on promising topics for dialogue between the US and China on AI ethics, safety, and governance
* [Dec 2024] Short paper on human-LLM interaction modes accepted as an oral in the Evaluating Evaluations (EvalEval) workshop at NeurIPS 2024
* [Sep 2024] Presented at Imperial College London's symposium on Human and Artificial Intelligence in Organizations
* [Jul 2024] New preprint on Open Technical Problems in AI Governance
* [Jun 2024] Presented "The Algorithm" at Sheffield DocFest
* [May 2024] Started internship with Google DeepMind's Ethics Research Team, working on model safety evaluation
* [May 2024] New preprint on human interaction evaluations for LLM safety, risks, and harms
* [Apr 2024] New preprint on harms from AI user interface designs, presented in CHI 2024 Workshop on Human-centered Evaluation and Auditing of Language Models
* [Feb 2024] Launched "The Algorithm" - a web-based game explainer of recommendation systems
* [Jan 2024] Joined Centre for Governance of AI as a winter research fellow to work on contextually-relevant model safety evaluations
* [Nov 2023] Awarded Dieter Schwarz Foundation-OII grant for research on AI, Government and Policy
* [Nov 2023] Presented work-in-progress game at International Documentary Film Festival of Amsterdam DocLab
See more
contact
lujainmibrahim@gmail.com
lujainmibrahim 
lujainibrahim