Research and product · Interpretability

Trenton Bricken

Designed his own Duke major in biological and artificial intelligence, then did a Harvard PhD in systems biology.

Career

Anthropic
- Member of Technical Staff, interpretability
Redwood Center for Theoretical Neuroscience, UC Berkeley
- Visiting researcher
Harvard Medical School, Debora Marks lab
- PhD research in systems biology
Duke University, Michael Lynch lab
- Undergraduate research; graduated 2020

PhD in Systems Biology from Harvard
self-made major in "Minds and Machines: Biological and Artificial Intelligence" from Duke University in May 2020

Was a visiting researcher at Berkeley's Redwood Center for Theoretical Neuroscience.
Spent a year in Michael Lynch's Duke lab using machine learning to design safer, more effective CRISPR guide RNAs.
Attended Duke as a Robertson Scholar, with full funding for all four years.
Wrote his Harvard thesis on sparse representations in biological and artificial neural networks.
Showed in a NeurIPS 2021 paper that transformer attention closely approximates Kanerva's Sparse Distributed Memory, a biologically plausible memory model.
Co-developed agents that audit models for misalignment; running many in parallel lifted one agent's solve rate from 13% to 42%.
Shoots analog film photography and keeps a site for his film photos.

Cross-Architecture Model Diffing with Crosscoders: Unsupervised Discovery of Differences Between LLMs (2026)
Findings from a Pilot Anthropic—OpenAI Alignment Evaluation Exercise (2025)
Building and evaluating alignment auditing agents (2025)
On the Biology of a Large Language Model (2025)
Circuit Tracing: Revealing Computational Graphs in Language Models (2025)
Auditing Language Models for Hidden Objectives (2025)