Kajetan Janiak (publicly known as Jett Janiak) is a mechanistic interpretability researcher from Poland who studied at the University of Warsaw, where he completed an MS in the Faculty of Mathematics, Informatics and Mechanics. He participated in the MATS (ML Alignment & Theory Scholars) Winter 2023 cohort under Neel Nanda's mentorship and later in a subsequent MATS cohort under Arthur Conmy, both in the mechanistic interpretability stream. His research focuses on understanding the internal mechanisms of transformer models, including work on polysemantic attention heads, circuit discovery in small transformers, stable regions in the residual stream of LLMs, sparse autoencoders, and chain-of-thought faithfulness in frontier models. He has co-authored several papers and Alignment Forum posts, including "Chain-of-Thought Reasoning In The Wild Is Not Always Faithful" (ICLR 2025 workshop) and "Characterizing Stable Regions in the Residual Stream of LLMs." He has also been involved with AI Safety Camp as a project lead. He received a grant from the Long-Term Future Fund to cover costs of leaving employment in order to pursue AI safety research.

Co-founder of Effective Giving and Ergo Impact, Dr. Kellie Liket is a social impact scholar who serves on the Development Cooperation Committee of the Netherlands Advisory Council on International Affairs (AIV) and completed a PhD on social impact measurement at Erasmus University after earning a master’s degree in development economics from the London School of Economics.

Horizon is a non-partisan nonprofit that addresses the US government's critical shortage of emerging technology expertise by recruiting, training, and placing technical talent in federal agencies, congressional offices, and think tanks.

James is the Chief Operations Officer at Giving What We Can, leading finance, compliance, people, grantmaking and operations across GWWC’s entities. He previously worked as a strategy and impact consultant supporting social impact organisations across Europe and the United States, and served as Head of Finance and Operations at Safe Passage Project, a nonprofit law firm serving refugee children in New York. He has a background in social and political philosophy and is completing a PhD in organisational science focused on improving organisations that serve vulnerable populations.

No summary available yet.

AI Security Forum and Hardware Lead at Heron, a non-profit AI security initiative based in Tel Aviv that connects cybersecurity experts with frontier AI security challenges.

No summary available yet.

Berlin-based AI safety research engineer who evaluates frontier AI models for dangerous capabilities and currently works with METR, EquiStamp, and Redwood Research on benchmarks and evaluations.

Policy and law background, will add more shortly.

Head of 5050 at Fifty Years, where she designs and runs the 5050 program end-to-end—from operations and recruiting to partner-led workshops, founder fireside chats, and retreats. Previously she led Sigueme, building houses for low-income families in Lima, co-founded the education nonprofit Enseñame to deliver online tutoring to thousands of students, and studied entrepreneurship at Wharton to scale her impact.

Philip Trippenbach is a communications strategist and Strategy Director at the Seismic Foundation, a non-profit focused on mobilizing public demand for responsible AI governance. He is based in Stockholm, Sweden and has over two decades of experience in communications, digital strategy, and AI policy. Prior to Seismic, he held senior roles at Edelman in London, including Head of Influence and Client Strategy Lead EMEA, where he invented the RARA framework of influence adopted by global clients. Earlier in his career he was a digital producer with the BBC Current Affairs team and Editor-in-Chief of a networked journalism startup. His work connects AI governance with strategic communications, and he has received Long-Term Future Fund grants to run workshops on strategic communications for the AI safety community. He co-authored the 2025 paper "From Catastrophic to Concrete: Reframing AI Risk Communication for Public Mobilization" (arXiv:2511.06525), examining how public engagement with AI risk shifts based on framing and messaging strategies.

Dr. Dileep George is Head of AI for Astera Institute’s Neuro & AGI program, leading its neuro‑inspired AGI research division. An entrepreneur, scientist, and engineer working at the intersection of AI, robotics, and neuroscience, he previously co‑founded the AI companies Numenta and Vicarious, co‑developed the Hierarchical Temporal Memory framework, and worked at Google DeepMind on agents with memory, planning, and structure learning. He holds an MS and PhD in Electrical Engineering from Stanford University and a B.Tech in Electrical Engineering from IIT Bombay.

Narayan Subramanian is a nonresident scholar in the Sustainability, Climate, and Geopolitics Program at the Carnegie Endowment for International Peace, focusing on clean‑energy finance and industrial strategy, after previously serving as director for energy transition on the White House National Security Council and as an advisor at the U.S. Department of Energy.

No summary available yet.

No summary available yet.

CNAS is a Washington, DC-based bipartisan think tank that develops national security and defense policy, with a dedicated Technology & National Security program focused on AI, compute governance, and great power competition.

No summary available yet.

No summary available yet.

No summary available yet.

Research Associate and Head of Training Programmes at ILINA, focusing on governance of frontier AI in Global South countries, especially monitoring and reporting AI harms; she has consulted for the UN Working Group on Business and Human Rights on AI issues, is involved with the AI Safety Student Team at Harvard, and holds a first‑class undergraduate law degree from Strathmore University and an LLM from Harvard Law School.

No summary available yet.

No summary available yet.

No summary available yet.

William MacAskill is a moral philosopher and Senior Research Fellow at Forethought whose work focuses on AGI preparedness and the long-term future. Before joining Forethought he was an Associate Professor of Philosophy at the University of Oxford. He is best known for co-founding the effective altruism movement through organisations such as Giving What We Can, 80,000 Hours, the Centre for Effective Altruism and the Global Priorities Institute, and for authoring or co-authoring the books Doing Good Better, What We Owe The Future, Moral Uncertainty and An Introduction to Utilitarianism.

Lucy Farnik is a PhD student at the University of Bristol (2023 cohort) in the UKRI Centre for Doctoral Training in Interactive Artificial Intelligence, supervised by Dr Conor Houghton and Mengyue Yang, with her research titled "Towards interpretable and controllable deep language modeling." She completed ARENA and the MATS research program under the mentorship of Neel Nanda at Google DeepMind, during which she explored SAE-based circuit-style analysis — work that led to an LTFF extension grant. Her research focuses on mechanistic interpretability of large language models, particularly sparse autoencoders (SAEs), with publications including "Residual Stream Analysis with Multi-Layer SAEs" (ICLR 2025) and "Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations" (ICML 2025). She has also been affiliated with FAR.AI and co-founded BAISC (Bristol AI Safety Student Community), a student research centre focused on AI safety. Her background includes a BEng in Computer Science with Innovation from the University of Bristol and extensive software engineering experience starting from an early age.

Agatha Duzan is a master’s student in Data Science at EPFL who works on AI safety. Public profiles and EPFL communications describe her as president of Safe AI Lausanne, the EPFL student association on AI safety, and her own activity notes that she co-led an AI safety bootcamp where she designed and taught the AI governance curriculum. She also contributes to AI safety research projects, including work on sparse autoencoders for unlearning dangerous capabilities and AI-assisted STEM education.

No summary available yet.

Hugo Larochelle is Scientific Director of Mila – Quebec Artificial Intelligence Institute, an adjunct professor at Université de Montréal, and a leading deep‑learning researcher. Previously, he led Google’s AI research lab in Montréal, and his work has helped shape modern deep‑learning systems used in industry and academia.

No summary available yet.

Part-time research scientist with Truthful AI; holds a PhD in Automatic Control and Robotics and works as an assistant professor at the Faculty of Mechatronics, Warsaw University of Technology; collaborates on AI safety projects related to out-of-context reasoning in large language models.

No summary available yet.

Swadeep Singh is General Manager – Data Science at IndiaAI, where he leads the data science function within the Government of India’s national AI mission.

No summary available yet.

No summary available yet.

Co-founder of Impact Ops and experienced operations professional; previously served as Head of Staff Support at Effective Ventures, Operations Specialist at the Centre for Effective Altruism, and Operations Manager at Brainlabs, and holds a first-class MA (Oxon) in Psychology and Philosophy from the University of Oxford.

Co‑founder and technology advisor at the Collective Intelligence Project, formerly a research engineer at DeepMind working on multi‑agent reinforcement learning and language models, and a researcher in technology governance with groups such as the Berkman Klein Center.

A private research university in Nashville, Tennessee, that received SFF Fairness Track funding for research related to AI fairness, algorithmic equity, and the societal implications of AI systems.

No summary available yet.

Codruta (Coco) Lugoj is an independent AI safety researcher based in London, UK. She holds a master's degree in AI and has a background in machine learning research, with particular expertise in reinforcement learning and probabilistic and variational inference, developed during her time at Radboud University. She received Long-Term Future Fund grants in 2023 and 2024 to build alignment research engineering skills and to continue her work evaluating agent self-improvement capabilities. As part of the Axiom Futures Alignment Research Fellowship, she co-authored "Auto-Enhance: Towards a Meta-Benchmark to Evaluate AI Agents' Ability to Improve Other Agents," presented at NeurIPS 2024, which proposes a meta-benchmark for measuring the ability of LLM agents to modify and improve other agents across tasks including prompt injection resiliency, dangerous knowledge unlearning, and solving real GitHub issues. Her research sits at the intersection of AI safety evaluation and agentic AI capabilities.

No summary available yet.

Dr Waku is a pseudonymous AI safety educator who creates YouTube videos, a Substack newsletter, and other content explaining AI alignment risks and AI security to general audiences.

Max Lamparth is a Research Fellow at the Hoover Institution's Technology Policy Accelerator at Stanford University, where he is also affiliated with the Stanford Intelligence Systems Laboratory and the Stanford Center for AI Safety. He holds a Ph.D. in Natural Sciences from the Technical University of Munich (2023) and B.Sc. and M.Sc. degrees in Physics from Ruprecht Karl University of Heidelberg. His research focuses on the security and safety of large language models through mechanistic interpretability, reward modeling, and robust evaluation, including developing model-internal interventions for failure modes such as backdoors and reward-proxy shortcuts. He created and teaches CS120: Introduction to AI Safety at Stanford, and his technical work has directly informed AI governance efforts through policy publications in outlets such as Foreign Affairs and the Carnegie Council for Ethics in International Affairs, as well as direct engagement with policymakers on AI legislation. His research has been published at NeurIPS, CoLM, FAccT, and AIES, and has received media coverage from MIT Technology Review, Axios, and New Scientist.

Viktoriya Malyasova is a machine learning engineer focused on AI alignment, based in the San Francisco Bay Area. She holds a Master's degree in Data Science from Skoltech (Skolkovo Institute of Science and Technology) and a Bachelor's degree in Mathematics from HSE (Higher School of Economics) in Russia. She has worked at FAR.AI (Frontier AI Alignment Research), a technical AI safety nonprofit, as well as at DeepSpin in Germany and MainsLab in Russia. She participated in the SERI MATS 2023 winter program as a scholar, where she focused on infrabayesianism — the mathematical framework developed by Vanessa Kosoy for reasoning under logical uncertainty and non-realizability. Prior to the SERI MATS program, she received retrospective funding for up-skilling in infrabayesianism, and she contributed to creating and testing educational exercises on inframeasure theory that were published on the AI Alignment Forum. She maintains a programming and machine learning blog at malyasova.github.io.

BlueDot Impact is a nonprofit talent accelerator that runs free cohort-based courses to train professionals in AI safety, AI governance, and biosecurity. It is the leading pipeline for building the workforce needed to safely navigate transformative AI.

No summary available yet.

A 501(c)(3) research laboratory in Santa Monica, CA that uses neuroimaging, neuromodulation, VR/AR, and altered states to study consciousness, with an AI safety research program on preventing antisocial AI through artificial empathy.

Xin (Eric) Wang is an Assistant Professor of Computer Science at UC Santa Barbara and Director of the UCSB Center for Responsible Machine Learning, with research focusing on multimodal and agentic AI.

No summary available yet.

EleutherAI is a nonprofit AI research institute focused on interpretability, alignment, and open-source foundation model research. It is best known for creating GPT-NeoX, the Pythia model suite, and The Pile dataset.