From Leaky Thoughts to Private Reasoning: Controlling What LRMs Say to Themselves
Published in arXiv preprint 2026, 2026
Haritz Puerto
Postdoctoral researcher at the ELLIS Institute Tübingen (COMPASS group), working on trustworthy, safe, and effective reasoning in language models and agentic systems.
- Tübingen, Germany
- ELLIS Institute Tübingen
- Google Scholar
- Semantic Scholar
- GitHub
- X (formerly Twitter)