Lukas Berglund
Bio
Lukas Berglund is an AI safety researcher currently serving as Technical Staff at the U.S. Center for AI Standards and Innovation (CAISI) at NIST. He is best known as the lead author of "The Reversal Curse: LLMs trained on 'A is B' fail to learn 'B is A'," published at ICLR 2024, which demonstrated a fundamental generalization failure in autoregressive large language models. He also co-authored "Taken out of context: On measuring situational awareness in LLMs," an influential paper exploring how models recognize whether they are in training or deployment. His research was conducted in part as a MATS Fellow through the SERI MATS program, with support from Open Philanthropy. He has an undergraduate background from Vanderbilt University and his work spans AI evaluation, AI security, and empirical research on the capabilities and failure modes of frontier AI systems.
Links
- Personal Website
- https://lukasberglund.com/
- Twitter / X
- LessWrong
- -
Grants
from Long-Term Future Fund
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.
Details
- Last Updated
- Mar 22, 2026, 11:15 PM UTC
- Created
- Mar 20, 2026, 2:54 AM UTC