Pepa Atanasova

Pepa Atanasova

Assistant Professor (Tenure Track)

Department of Computer Science

CopeNLU Group

I am an Assistant Professor (Tenure Track) in the Department of Computer Science, AI Section at the University of Copenhagen. I co-lead the CopeNLU group.

Research

Interpretability of Language Models

Understanding mechanisms of LLMs with applications to context usage and parametric knowledge.

Explainability Methods

Designing robust and user-aligned explainability techniques for complex models.

Factuality in LMs

Addressing the challenge of maintaining factual accuracy in language models.

Recent News

September 2026 Awarded Amazon Research Awards 2026 in Information Security for Security-Relevant Chain-of-Thought Steering for Agent Monitoring! Will be looking to hire a research assistant to work on the project.
August 2026 Honored to receive the H.C. Ørsted Forskertalentprisen 2026 for research into more trustworthy, transparent and responsible AI language models.
June 2026 Convergence Grant: Received funding from the University of Copenhagen for "DeceptionWeb: Foundations for Emergent Deception in Human-AI Interaction" at the intersection of NLP, Psychology, and Economic Game-theory, with Ingo Zettler (SODAS) and Egor Starkov (ECON).
May 2026 Received DFF-Research Project 2 grant for "A Mechanistic Framework for Mitigating the Susceptibility of LLMs to Learning False Information". Co-PI with Isabelle Augenstein.
May 2026 Selected to serve as Publication Chair for EACL 2027.
March 2026 Selected to serve as Secretary to SIGREP 2026-2028.
March 2026 Selected to serve as Industry Track Chair for EMNLP 2026.
March 2026 Invited talk at Aurora Huawei Summit in Helsinki "When Explanations Lie" at the LLM Safety Workshop [Slides]
December 2025 Invited talk at ELLIS Unconference "When Explanations Lie" at the LLM Safety Workshop [Slides]
November 2025 Talk at Digital Tech Summit on "Reality Check of LLM-driven Fact Verification" [Slides]