Team

Owain Evans

Director and Research Lead

Owain is also an affiliate at CHAI (UC Berkeley) and was previously based at the University of Oxford at the Future of Humanity Institute. He earned his PhD at MIT. He also worked at Ought, where he still serves on the Board of Directors.

Website | Google Scholar | 𝕏 @OwainEvans_UK | Wikipedia

Jan Betley

Research Scientist

Jan worked as a software developer for over a decade before shifting to AI safety in 2023. His work has focused on out-of-context reasoning, unexpected generalization from finetuning, and persona steering in LLMs. He was lead author of Emergent Misalignment, published in Nature in 2026, and most recently of Value Leakage, which shows that a model's responses can be silently shaped by its own values. He is an ARENA and Astra Fellowship alumnus.

Website | Google Scholar | 𝕏 @BetleyJan

Jorio Cocola

Research Scientist

Jorio has a PhD in mathematics from Northeastern University and was a postdoc in applied mathematics at Harvard. He worked on random matrix theory and the mathematics of deep learning before moving into AI safety as a MATS scholar. His recent work has focused on how finetuning shapes a model's persona. He led Story Imprinting, which shows that AI assistants can take on traits from human characters they resemble, and co-authored Weird Generalization & Inductive Backdoors.

Google Scholar | 𝕏 @JorioCocola

Jan Dubiński

Research Scientist

Jan has a PhD in Computer Science from the Warsaw University of Technology and is an Astra Fellowship and MARS alumnus. He is interested in AI safety, particularly in understanding and evaluating failure modes in generative models.

Website | Google Scholar | 𝕏 @jan_dubinski_

Adam Karvonen

Research Scientist

Adam's work focuses on explaining LLM behavior through interpretability and model self-explanation. He was lead author on Activation Oracles, an oral paper at ICML 2026, which trains LLMs to answer plain-language questions about another model's internal activations. He also led CHIVE, a pipeline for discovering in-the-wild LLM behaviors with unexpected causes, and SAEBench (ICML 2025), a benchmark for sparse autoencoders. He is a MATS alumnus and joined Truthful AI in October 2026.

Website | Google Scholar | 𝕏 @a_karvonen

Anna Sztyber-Betley

Research Scientist (part-time)

Anna has a PhD in Automatic Control and Robotics and is an assistant professor at the Faculty of Mechatronics, Warsaw University of Technology. She collaborates with Truthful AI on AI safety projects related to out-of-context reasoning in LLMs.

Google Scholar | 𝕏 @anna_sztyber

Johannes Treutlein

Research Scientist

Johannes was previously a Member of Technical Staff on Anthropic's alignment team, where he worked on alignment auditing, including Auditing Language Models for Hidden Objectives. He is on leave from a PhD in computer science at UC Berkeley, supervised by Stuart Russell. His recent work has focused on motivated reasoning in language models, such as Value Leakage. He was also lead author of Connecting the Dots (NeurIPS 2024).

Website | Google Scholar | 𝕏 @j_treutlein

Karol Gałązka

Research Engineer

Karol is an experienced software engineer and entrepreneur. In 2025, he participated in ML4Good and began working on AI safety projects.

Google Scholar | 𝕏 @karolgalazka

Nate Nicholson

Director of Operations

Nate has an extensive background in nonprofit operations, leadership, and development. He joined Truthful AI in September 2026.

LinkedIn

Alumni

James Chua

Research Scientist (Alumnus)

James was the second employee at Truthful AI and contributed to many projects. He is now a research scientist in alignment science at Anthropic.

Website | Google Scholar | 𝕏 @jameschua_sg

Current and Past Mentees

Owain Evans has mentored more than 40 researchers, many of whom have gone on to roles across the AI safety field, including full-time research positions at Truthful AI. If you are interested in working as a mentee, please consider applying to the Astra Fellowship or MATS.