Papers

For a full list of publications, see my Google Scholar.

Selected papers

Real-Time Detection of Hallucinated Entities in Long-Form Generation
Oscar Obeso*, Andy Arditi*, Javier Ferrando, Joshua Freeman, Cameron Holmes, Neel Nanda
arXiv preprint arXiv:2509.03531, 2025
Persona Vectors: Monitoring and Controlling Character Traits in Language Models
Runjin Chen, Andy Arditi, Henry Sleight, Owain Evans, Jack Lindsey
arXiv preprint arXiv:2507.21509, 2025
Refusal in Language Models Is Mediated by a Single Direction
Andy Arditi*, Oscar Obeso*, Aaquib Syed, Daniel Paleka, Nina Panickssery, Wes Gurnee, Neel Nanda
Advances in Neural Information Processing Systems, 2024

Other papers

Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs
Jan Betley, Jorio Cocola, Dylan Feng, James Chua, Andy Arditi, Anna Sztyber-Betley, Owain Evans
arXiv preprint arXiv:2512.09742, 2025
Inverse Scaling in Test-Time Compute
Aryo Pradipta Gema, Alexander Hägele, Runjin Chen, Andy Arditi, Jacob Goldman-Wetzler, Kit Fraser-Taliente, Henry Sleight, Linda Petrini, Julian Michael, Beatrice Alex, Pasquale Minervini, Yanda Chen, Joe Benton, Ethan Perez
Transactions on Machine Learning Research, 2025
Where Do Reasoning Models Refuse?
Kureha Yamaguchi, Benjamin Etheridge, Andy Arditi
arXiv preprint arXiv:2507.03167, 2025
Teaching Models to Verbalize Reward Hacking in Chain-of-Thought Reasoning
Miles Turpin, Andy Arditi, Marvin Li, Joe Benton, Julian Michael
arXiv preprint arXiv:2506.22777, 2025
Structural Causal Bandits under Markov Equivalence
Min Woo Park, Andy Arditi, Elias Bareinboim, Sanghack Lee
Advances in Neural Information Processing Systems, 2025

Research blog posts

In-context learning of representations can be explained by induction circuits
Andy Arditi
ICLR Blogposts Track, 2026
Finding "misaligned persona" features in open-weight models
Andy Arditi, Runjin Chen
LessWrong, 2025
Follow-up experiments on preventative steering
Runjin Chen, Andy Arditi
LessWrong, 2025
Do models say what they learn?
Andy Arditi, Marvin Li, Joe Benton, Miles Turpin
LessWrong, 2025
Finding features causally upstream of refusal
Daniel Lee, Eric Breck, Andy Arditi
LessWrong, 2025
Unlearning via RMU is mostly shallow
Andy Arditi, Bilal Chughtai
LessWrong, 2024
Refusal mechanisms: initial experiments with Llama-2-7b-chat
Andy Arditi, Oscar Obeso
LessWrong, 2023

Talks

A mechanistic view of refusal in language models
Andy Arditi
Machine Learning Security Seminar Series (University of Cagliari), 2025