The ELLIS Institute Tübingen is pleased to welcome Dr. Nouha Dziri as an Adjunct Principal Investigator. Nouha, an expert in LLM reasoning, security & safety, and multi-agent systems for large language models, joins the Institute while continuing her role as a Senior Research Scientist at Cohere and an Associate Industry Member of Mila – Quebec AI Institute. Her affiliation strengthens the Institute’s work at the intersection of foundational machine learning research and frontier industry practice.
Nouha’s research seeks to advance the foundations of capable, reliable, and secure AI agents. She develops post-training methods and studies the training dynamics that enable language models to reason, generalize, and coordinate as collective systems to solve complex, long-horizon problems. Security and safety are central to her research, spanning the use of autonomous agents to discover and exploit vulnerabilities, strengthen cyber defenses, as well as securing the models themselves against attacks and manipulation.
Nouha brings a distinguished research background spanning leading academic and industry labs:
- She earned her PhD from the University of Alberta and the Alberta Machine Intelligence Institute (Amii) in 2023.
- She was a Postdoctoral Researcher with Prof. Yejin Choi at the Allen Institute for AI (Ai2), and went on to co-lead the post-training effort behind the open language model OLMo.
- Earlier in her career she interned at Google DeepMind, Microsoft Research, and Mila.
- She is the recipient of multiple honors, including Best Paper Awards at NeurIPS 2025 and NAACL 2025.
- Her work has been featured in leading outlets including The Economist, TechCrunch, Le Monde, Science, and Quanta Magazine.
Recognized Track Record in Reasoning and AI Safety&Security
Nouha’s contributions have been widely recognized by the global research community. She received a Best Paper Award at NeurIPS 2025 and a Best Paper Award at NAACL 2025. Her work investigated fundamental questions about reasoning in transformers in Faith and Fate: Limits of Transformers on Compositionality (NeurIPS 2023 Spotlight), a work that sparked numerous discussions in the community and became a widely cited reference in the debate about how far LLMs can actually reason. She is also a strong advocate for AI safety and security, and led the creation of a comprehensive safety suite for OLMo, including the automated red-teaming framework WildTeaming and the safeguarding system WildGuard (both NeurIPS 2024). Her research has been featured in Quanta Magazine, TechCrunch, Science, Le Monde, and Science News, reflecting its relevance well beyond the machine learning community.
Expanding the Institute’s Impact
By joining as an Adjunct Principal Investigator, Nouha reinforces the ELLIS Institute Tübingen’s mission of advancing world-class machine learning research. Her expertise in reasoning, safety, and multi-agent intelligence ensures Tübingen remains a focal point for the responsible development of increasingly capable AI.
→ Read more about the ELLIS Institute research groups