@ University of Oxford
Led by Prof. Adam Mahdi, the lab advances the science of AI evaluation, benchmarking, safety and security. Through rigorous empirical research we study how LLMs and agentic systems reason, interact with humans and drive scientific discovery — alongside industry partners deploying AI where reliability matters.
We treat language models the way a public laboratory treats any new instrument: measure first, theorise second, publish the full method so others can check the work.
We develop the science of LLM evaluation, setting the standard for rigorous assessment and identifying hidden risks before they matter.
RecentStrategic Navigation or Stochastic Search? How Agents and Humans Reason Over Document CollectionsFrom bias and toxicity to agentic misalignment, we study the full spectrum of AI risk and develop the technical and governance tools to address it.
RecentA Positive Case for Faithfulness: LLM Self-Explanations Help Predict Model BehaviorWe build agentic systems that automate scientific knowledge synthesis and discovery, with a focus on agents that are reliable, transparent and domain-grounded.
RecentStrategic Navigation or Stochastic Search? How Agents and Humans Reason Over Document CollectionsWe run large-scale empirical studies on how people use AI for high stakes decisions, from healthcare and law to policy and beyond.
The same systems now answering medical questions, drafting policy, and mediating public information should be measured with the same care as any other tool that affects public life.
A benchmark that tells real navigation apart from stochastic search when agents work over document collections.
Ł Borchmann, J Van Landeghem, M Turski, S Padarha, RO Kearns, A Mahdi, et al.
Peer-reviewed papers at ICML, NeurIPS, ICLR, EMNLP, and Nature Medicine. The lab keeps no private benchmarks. Read the work.
A complete bibliography, including pre-prints and unreviewed working papers, is published on the lab's GitHub.
See full bibliography
Adam leads OxRML. The group studies how language models reason, how people work with them, and how agentic systems behave on real scientific and decision-making tasks.
Felix KronesDPhil StudentMultimodal AI, digital health
Djavan De ClercqDPhil StudentAI and food security, LLMs
Andrew M. BeanDPhil StudentLLM evaluations, human–LLM interaction
Yushi YangDPhil StudentLLM & agentic post-training, AI alignment
Harry MayneDPhil StudentLLM interpretability, AI safety, LLM evaluations
Jessica RodriguesDPhil StudentKnowledge graphs, metascience
Guy ParsonsDPhil StudentHealthcare AI, digital health
Karolina KorgulDPhil StudentAI safety, agentic AI
Ryan Othniel KearnsDPhil StudentScience of evals, reasoning in LLMs
Shreyansh PadarhaDPhil StudentAI for science, AI safety, LLM evaluations
Mia KussmanMSc StudentHuman–LLM interaction, LLM evaluations
Caleb TanMSc StudentLLM evaluations, reasoning
Sebastian PetricVisiting Policy FellowLLMs and financial time series
Tristan NaidooResearch AffiliatePublic health AI, LLM evaluations
Josh LawmanEntrepreneur in ResidenceResearch-to-product translationThree ways in. Each begins with a conversation, and each ends with published outputs the public can read. We do not run NDAs over findings; we work on questions that benefit from being in the open.
On-site sessions for product and ML teams on evaluation, safety, and agent reliability.
Half-day to multi-week formats. For teams shipping LLM products in healthcare, finance, retail, and government.
We work with engineering partners to turn lab work into tools other teams can run.
Evaluation harnesses, safety dashboards, agentic-research platforms. We build them with partners we trust, carrying the research methods through to the code.
Applied research collaborations with foundations, governments, and large companies.
Multi-year programmes: shared roadmaps, sponsored DPhil studentships, named labs.
