Egor Zverev
Bio
Egor Zverev is a PhD student at the Institute of Science and Technology Austria (ISTA), supervised by Christoph Lampert and co-supervised by Florian Tramèr at ETH Zürich through the ELLIS PhD Program. He holds a B.S. with Honors in Applied Mathematics and Computer Science from the Yandex Department of Data Analysis at the Moscow Institute of Physics and Technology (MIPT). His research focuses on AI safety and security, with a particular emphasis on improving large language model security through architectural approaches. He is best known for formalizing the problem of instruction-data separation in LLMs and for developing ASIDE, an architectural method that separates instructions from data at the token embedding level using orthogonal rotations. His work has been published at ICLR 2025 and ICLR 2026, and he co-organized the LLM Safety and Security Workshop at EurIPS 2025. He received a relocation grant from the Long-Term Future Fund to support his move into the PhD program.
Links
- Personal Website
- https://egozverev.github.io/
- Twitter / X
- LessWrong
- egor-zverev
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, 3:45 PM UTC
- Created
- Mar 20, 2026, 2:50 AM UTC