👋 About
I am a M.Sc. student in Statistics and Data Science (Machine Learning track) at LMU Munich and a fellow of the Konrad Zuse School of Excellence in Reliable AI (relAI). My research is about building trustworthy AI systems - I work on AI safety, uncertainty quantification, and monitoring LLMs through their internal representations.
Currently, I am a research assistant at the SODA Lab (LMU Munich), where I investigate how moral foundations contribute to the robustness of refusal mechanisms in LLMs, and I work on deception detection in LLMs at the Data Analytics and Machine Learning group (TUM).
Before graduate school, I was a data scientist at Kept (ex-KPMG), building RAG and information-security systems for the security team, and a solution engineer intern at Toloka.AI (Yandex), working on fraud detection. I am always happy to chat about research - feel free to reach out at sofiia.nikolenko@campus.lmu.de.
🔬 Research Interests
Detecting and predicting unsafe behavior in LLMs — jailbreaks, deception, goal drift — before it turns into harm.
Reading safety signals from a model's internal dynamics rather than its outputs alone.
Uncertainty as a safety signal — when models should abstain, and when their monitors should be trusted.
📰 News
- Jul 2026 🏛️ Started as a research assistant at the Data Analytics and Machine Learning Group at TUM, working on AI safety.
- Jul 2026 🧩 Contributed vision-language explanation support to shapiq — see the demo.
- Jun 2026 🎉 Our paper What Intermediate Layers Know was accepted at ECML PKDD 2026, with a short version at the EIML workshop @ ICML 2026!
- Dec 2025 🔬 Joined the SODA Lab at LMU Munich as a research assistant, working on AI safety through LLM alignment.
- Oct 2024 🎓 Selected as a fellow of the Konrad Zuse School of Excellence in Reliable AI.
📄 Publications
BibTeX
@inproceedings{nikolenko2026intermediate,
title = {What Intermediate Layers Know: Detecting Jailbreaks
from Entropy Dynamics},
author = {Nikolenko, Sofiia and Papucci, Michele and
Rezaei, Mina and Manchingal, Shireen Kudukkil},
booktitle = {European Conference on Machine Learning and Principles
and Practice of Knowledge Discovery in Databases
(ECML PKDD)},
year = {2026},
eprint = {2606.25182},
archivePrefix = {arXiv}
}