# Chris Cundy **Contact:** chris.j.cundy@gmail.com **GitHub:** [C-J-Cundy](https://github.com/C-J-Cundy) --- ## Experience ### Research Scientist **FAR.AI** | Berkeley, California, USA | June 2024–Present Characterising and mitigating catastrophic risks from frontier AI systems. Main responsibilities: - Leading research projects and determining the direction for teams of researchers and engineers - Designing and implementing training and evaluation schemes with state-of-the-art models, conducting empirical studies, and communicating results through papers and presentations - As a research team lead, ensuring that research aims are consistent with FAR's mission and that my direct reports are supported in their output and professional development ### Research Scientist Intern **Technical AI Safety Team, DeepMind** | London, UK | June 2022–September 2022 Investigating robust and reliable machine learning in theory and at scale: - Investigated susceptibility of autoregressive models to 'delusions', where unobserved latent variables lead to incorrect probabilistic judgments - Developed a theoretical model for delusions; investigated delusions at scale by analysing performance of DeepMind's Gato (a large generalist, multi-task autoregressive model) on custom environments ### Visiting Scholar **Future of Humanity Institute, University of Oxford** | Oxford, UK | October 2017–January 2018 Developing algorithms to predict human judgments. Supervised by Owain Evans and Andreas Stuhlmüller: - Designed algorithms to collate quick, noisy human judgments to predict the answer to complicated tasks which would typically require deliberation ### Visiting Scholar **Centre for Human-Compatible AI, University of California, Berkeley** | US | June–September 2017 Supervised by Daniel Filan & Stuart Russell, researching topics in AI safety: - Extended previous work on inverse reinforcement learning to hierarchical setting. Formalized the problem, derived theoretical results, performed experiments on data and presented at an ICML workshop --- ## Education ### PhD - Computer Science **Stanford University** | Stanford, California, USA | 2018–2024 - Advised by Stefano Ermon - Investigating topics in inverse reinforcement learning, sequence modelling and variational inference - Thesis: *Beyond Maximum Likelihood: Distribution-Aware Machine Learning* ### MEng - Computer Science **University of Cambridge** | Cambridge, UK | 2016–2017 - Grade: Distinction - Modules: Data Science, Probabilistic Machine Learning, Network Analytics - Thesis: *Investigating Variational Gaussian Process State-Space Models with Gaussian Likelihood*. Supervised by Carl E. Rasmussen ### BA - Natural Sciences (Physics) **University of Cambridge** | Cambridge, UK | 2013–2016 - Grade: 1st - Modules: Physics, Maths, Chemistry, Computer Science --- ## Selected Publications **The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception Probes** Mohammad Taufeeque, Stefan Heimersheim, Adam Gleave, Chris Cundy | Preprint, 2026 **Auditing Games for Sandbagging** Jordan Taylor, Sid Black, Dillon Bowen, Thomas Read, Satvik Golechha, Alex Zelenka-Martin, Oliver Makins, Connor Kissane, Kola Ayonrinde, Jacob Merizian, Samuel Marks, Chris Cundy, Joseph Bloom | Technical Report, 2025 **Preference Learning with Lie Detectors can Induce Honesty or Evasion** Chris Cundy, Adam Gleave | NeurIPS 2025 **SequenceMatch: Imitation Learning for Autoregressive Sequence Modelling with Backtracking** Chris Cundy, Stefano Ermon | ICLR 2024 **Privacy-Constrained Policies via Mutual Information Regularized Policy Gradients** Chris Cundy, Rishi Desai, Stefano Ermon | AISTATS 2024 **LMPriors: Pre-Trained Language Models as Task-Specific Priors** Kristy Choi*, Chris Cundy*, Sanjari Srivasta, Stefano Ermon | First Workshop on Foundation Models for Decision Making, NeurIPS 2022 **BCD Nets: Scalable Variational Approaches for Bayesian Causal Discovery** Chris Cundy, Aditya Grover, Stefano Ermon | NeurIPS 2021 **Flexible Approximate Inference via Stratified Normalizing Flows** Chris Cundy, Stefano Ermon | UAI 2020 **Parallelizing Linear Recurrent Neural Nets over Sequence Length** Eric Martin, Chris Cundy | ICLR 2018 --- ## Additional Publications **Sharpe Ratio-Guided Active Learning for Preference Optimization in RLHF** Syrine Belakaria, Joshua Kazdan, Charles Marx, Chris Cundy, Willie Neiswanger, Sanmi Koyejo, Barbara E Engelhardt, Stefano Ermon | CoLM 2025 **A physics-informed machine learning model for the prediction of drop breakup in two-phase flows** Chris Cundy, Shahab Mirjalili, Charlélie Laurent, Stefano Ermon, Gianluca Iaccarino, Ali Mani | International Journal of Multiphase Flow, 2024 **Neural Networks and the Chomsky Hierarchy** Grégoire Delétang, Anian Ruoss, Jordi Grau-Moya, Tim Genewein, Li Kevin Wenliang, Elliot Catt, Chris Cundy, Marcus Hutter, Shane Legg, Joel Veness, Pedro A. Ortega | ICLR 2023 **Towards a foundation model for geospatial artificial intelligence** Gengchen Mai, Chris Cundy, Kristy Choi, Yingjie Hu, Ni Lao, Stefano Ermon | Proceedings of the 30th International Conference on Advances in Geographic Information Systems, 2022 **IQ-Learn: Inverse soft-Q Learning for Imitation** Divyansh Garg, Shuvam Chakraborty, Chris Cundy, Jiaming Song, Stefano Ermon | NeurIPS 2021 **Exploring Hierarchy-Aware Inverse Reinforcement Learning** Chris Cundy, Daniel Filan | Workshop on Goal Specifications for Reinforcement Learning, ICML 2018 **Predicting Slow Judgment** Owain Evans, Andreas Stuhlmüller, Ryan Carey, Neal Jean, Andrew Schreiber, Girish Sastry, Chris Cundy | Aligned Artificial Intelligence Workshop, NeurIPS 2017 --- ## Service **Participant, EU AI Act Code of Practice Working Groups 2 and 4** | 2025 Participated, as an independent expert, in working groups 2 and 4 for the development of the EU AI Act Code of Practice (CoP). I advocated, via written and oral presentation, for the importance of pre- and post-mitigation model evaluations, outlined in an earlier position paper I authored. **Teaching Assistant—CS228 (Probabilistic Graphical Models)** | Stanford University | 2023 **Head Teaching Assistant—CS228 (Probabilistic Graphical Models)** | Stanford University | 2022 Received award for excellence (awarded to top 5% of Teaching Assistants). **Project Supervisor** | Supervised Project for Alignment Research (SPAR), Stanford AI Alignment | 2023 Supervised five undergraduates on a project finding scaling laws in prompt injections. Presented work at the 7th Center for Human-Compatible AI workshop. **Project Supervisor** | Undergraduate Research Program, Stanford Existential Risk Initiative | 2021 Served as supervisor for an undergraduate project on forecasting AI progress. **Reviewer** | 2020–Present Reviewed for the following venues: UAI (2020, 2022, 2025, 2026), ICML (2019, 2020, 2023, 2025, 2026), ICLR (2021-2026), NeurIPS (2021-2025), AAAI (Safe and Robust AI track) (2023-2024), TMLR (2025, 2026). --- ## Relevant Awards **Winner, OpenAI Preparedness Challenge** | March 2024 - One of the top ten submissions for the OpenAI Preparedness Challenge, for submitting *the most unique, while still being probable, potentially catastrophic misuse of the [OpenAI API]* - Developed proof-of-concept showing how GPT4-V, and speech-to-text with GPT4, could be used to parse vast amounts of unlabelled surveillance data, finding actionable insights for blackmail or insider trading - Prize: $25,000 in OpenAI credits