Our Winter 2026 Fellows

Ashwin works on technical AI safety research. He was previously a software engineer working on machine learning compliance infrastructure at Google. Before this, he graduated from the University of Michigan with a degree in computer science. He has worked on projects in AI control, machine unlearning, computer vision for medical imaging, and natural language generation.

Girish is an AI researcher focused on understanding the internal mechanisms of large-scale AI systems and their implications for AI safety. His interdisciplinary background spans theoretical physics and a decade of investigative journalism covering global crises from Venezuela to Iraq, with everyone from the New Yorker to Reuters. Most recently, he has led AI and software engineering in San Francisco across human rights investigations, accounting, and healthcare. He lives with his wife and children in San Francisco, and enjoys photography, music, and learning history.

Kyungeun holds a PhD in Physics from Columbia University and conducted postdoctoral research at Yale University, co-authoring 25+ publications in international dark matter and neutrino physics collaborations. She spent recent years in industry ML/data science on both US coasts covering a big corporation and a startup. Her current research focuses on machine unlearning and interpretability, bringing experimental physics methodology to develop rigorous evaluation frameworks for AI safety. (https://kyungeunlim.github.io/)

Lennart is a technical AI safety fellow at ERA and a soon-to-graduate PhD candidate at the CDT in Safe and Trusted AI at King’s and Imperial College London. Previously, he interned at Google DeepMind and worked as an NLP researcher at the University of Vienna. His research spans explainability, human–AI interaction, and robotics.

Marek is interested in developing detection methods for model self-awareness as the first step towards securing model welfare and assuring a benevolent relationship between humans and digital minds. He holds a medical degree and a PhD in computational neuroscience from Boston University, where he investigated neural mechanisms implicated in altered consciousness under ketamine anesthesia. Most recently, he has been practicing internal medicine as a junior faculty at the UCLA David Geffen School of Medicine.

Matthew founded the AI Safety and Governance Group at Northwestern University, where he's completing his Master's in Computer Science. Before ERA, Matthew worked as a software engineer for Amazon and Northwestern's Global Poverty Research Lab. He has researched LLM interpretability, automated research, and AI control. In his free time, he enjoys designing video games and solving escape rooms.

Nathaniel holds a BSc in Mathematics and a BEng in Data Science and Engineering, and is an AI researcher with experience across industry and academia. He has participated in SPAR, working on neural scaling laws grounded in data distribution geometry alongside Ari Brill, and at AISC, where he focused on preventing adversarial reward optimization with Domenic Rosati. He has also interned at Geodesic Research, studying obfuscation generalization with Cam Tice and Puria Radmard, and more recently at Qualcomm AI Research, where he led a project on multimodal and multilingual reasoning. Nathaniel is currently a visiting researcher at the van der Schaar Lab at the University of Cambridge, where his work centers on reasoning under epistemic uncertainty with applications to medical diagnostics.

Peter is an MSc Computer Science student at ETH Zurich with a BSc in Computational Mathematics. At ETH, he has worked with the LAS and SPY groups on anthropomorphized misalignment, focusing on deception detection in large language models. His current interests center on mechanistic interpretability research.

Scott is a cognitive scientist pivoting to technical AI safety, with a focus on building and validating evaluation frameworks for social cognition, deception precursors, and other epistemic failure modes in large language models. His work draws on human psychology—especially theory of mind, apophenia, and individual differences—to better characterize and mitigate risks like hallucination, manipulation, and strategic misbehavior in AI systems. He previously completed a postdoc at University of Michigan and earned his PhD in Psychology from University of Minnesota. Outside research, he’s an avid runner and pianist.

Andrew is a current undergraduate studying Computer Science and Public Policy at Georgia Tech. He is a lead organizer for the Georgia Tech AI Safety Initiative, through which he's presented at a Congressional exhibition, co-authored an RFI for the National AI Plan, and helped found governance field building efforts. His current research interests include multi-agent risks, particularly collusion and concentration of power, and improving how models generalize safety training, drawing on work in emergent misalignment and inoculation prompting.

Botao 'Amber' Hu is a PhD candidate in Human Centred Computing at the University of Oxford's Department of Computer Science. His research focuses on Decentralized AI (DeAI), exploring both governance and applications.

Henry is a former quantitative trader with a Bachelor’s degree in Mathematics from the University of Cambridge and a Master’s degree in Financial Mathematics from the Technical University of Munich. His current interests focus on the alignment problem, particularly the design and scaling of evaluations to detect misaligned behaviour. Outside of work, Henry trains for middle-distance triathlons and begins each day with a coffee and the New York Times crossword.

Jacob is an AI safety researcher at Exona Labs. Previously, he was a participant of ARENA 6.0, a researcher at a digital policy think tank based in the University of Cambridge on AI evaluation and policy, and an AI Linguist at LinkedIn. He graduated from the University of Edinburgh in 2024 with an MSc in Speech and Language Processing, and from the University of Cambridge with a BA in Computational Linguistics. His research interests include: AI monitoring frameworks, agentic AI governance, mechanistic interpretability, organisational AI risk quantification, and AI transparency standards.

Nick works in safeguards enforcement at Anthropic, where he evaluates catastrophic harm threats and develops safety protocols. Previously, he trained generative AI models at Meta and conducted research on AI for sustainable development at the United Nations Development Programme. He holds a Master of Theological Studies from Harvard Divinity School, focusing on religion, ethics, and politics. His career sits at the intersection of AI safety, policy, and global development.

Oliver works at the intersection of technical AI governance and AI-driven psychosocial impact. His research focuses on mitigating societal risks from emerging misalignment, improving model organisms/character training, and advancing global cooperation on AI governance. He’s currently building a CogSec startup backed by the BlueDot Incubator, and is a MARS AI Governance Fellow researching US export controls & US-China AI competition dynamics. Previously, Oliver worked as a Deeptech VC, was a Cambridge Enterprise fellow, and led ML-enabled research in biophysics. He holds an MPhil in Physics (University of Cambridge) and a BSc in Computational Biology & Mathematics (UCLA), and is currently on leave from his PhD at Columbia University.

Vilija is an AI safety researcher and entrepreneur. She combines consulting experience in Business Resilience (Deloitte) and Quantitative Solutions (PwC) with research expertise on AI, cybersecurity governance, and epistemic resilience, and a track record of leadership within the technology space (part of TNW T500). With a focus on human-centric AI-enhanced solutions, she is a co-founder and European Strategy Lead at Encode Europe, and AI governance non-profit.